{"asset": "gpu-cuda-pytorch-compatibility", "asset_type": "lookup", "body": "One record per (PyTorch release, CUDA build) pairing: which CUDA builds each PyTorch release ships wheels for, which Python versions it supports, which cuDNN it carries, which GPU architectures (compute capabilities) the wheel was compiled for, and the minimum NVIDIA driver the pairing needs. Answers 'does torch 2.7 run on an RTX 5090', 'nvidia-smi says CUDA 12.4 but torch wants 12.6 \u2014 do I need a new driver', 'which pip index URL do I use for CUDA 12.8', 'why is torch.cuda.is_available() False on a card that CUDA supports'. Nobody publishes this join: PyTorch's release notes state the CUDA builds but not the driver floor, NVIDIA's release notes state the driver floor but say nothing about PyTorch, and the compiled architecture list lives only in PyTorch's own build script. The pairing is what a person actually needs and it is the thing assistants get wrong.\n", "description": "One record per (PyTorch release, CUDA build) pairing: which CUDA builds each PyTorch release ships wheels for, which Python versions it supports, which cuDNN it carries, which GPU architectures (compute capabilities) the wheel was compiled for, and the minimum NVIDIA driver the pairing needs. Answers 'does torch 2.7 run on an RTX 5090', 'nvidia-smi says CUDA 12.4 but torch wants 12.6 \u2014 do I need a new driver', 'which pip index URL do I use for CUDA 12.8', 'why is torch.cuda.is_available() False on a card that CUDA supports'. Nobody publishes this join: PyTorch's release notes state the CUDA builds but not the driver floor, NVIDIA's release notes state the driver floor but say nothing about PyTorch, and the compiled architecture list lives only in PyTorch's own build script. The pairing is what a person actually needs and it is the thing assistants get wrong.", "file": "index.md", "generated": true, "harvested": "2026-09-16", "key_field": "pairing", "licence": "PyTorch's RELEASE.md and build scripts are BSD-3-Clause (pytorch/pytorch LICENSE); NVIDIA's CUDA release notes are vendor technical documentation. We take facts (version pairings, driver floors, architecture lists) plus a short attributed quote linking back to each source. We do not reproduce either document.", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/1.12/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/1.12/windows/cuda113.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html", "https://raw.githubusercontent.com/pytorch/builder/release/1.12/windows/cuda116.bat", "https://raw.githubusercontent.com/pytorch/builder/release/1.13/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/1.13/windows/cuda116.bat", "https://raw.githubusercontent.com/pytorch/builder/release/1.13/windows/cuda117.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.0/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.0/windows/cuda117.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.0/windows/cuda118.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.1/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.1/windows/cuda118.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.1/windows/cuda121.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/pytorch/windows/cuda126.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/pytorch/windows/cuda128.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/pytorch/windows/cuda130.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.11/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.11/.ci/pytorch/windows/internal/cuda_config.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.12/.ci/pytorch/windows/internal/cuda_config.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.13/.ci/pytorch/windows/internal/cuda_config.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.2/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.2/windows/cuda118.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.2/windows/cuda121.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.3/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.3/windows/cuda118.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.3/windows/cuda121.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/windows/cuda118.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/windows/cuda121.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/windows/cuda124.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/windows/cuda118.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/windows/cuda121.bat", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/windows/cuda124.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/pytorch/windows/cuda118.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/pytorch/windows/cuda124.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/pytorch/windows/cuda126.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/pytorch/windows/cuda118.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/pytorch/windows/cuda126.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/pytorch/windows/cuda128.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/pytorch/windows/cuda126.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/pytorch/windows/cuda128.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/pytorch/windows/cuda129.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/pytorch/windows/cuda126.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/pytorch/windows/cuda128.bat", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/pytorch/windows/cuda130.bat"], "stale_after": "2026-12-15", "title": "PyTorch release, CUDA build and NVIDIA driver pairings", "type": "dataset", "verified": false} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 1.12 with CUDA 11.3\n\n**PyTorch release:** 1.12\n\n**CUDA build:** CUDA 11.3\n\n**Python versions:** >=3.7, <=3.10\n\n**cuDNN version:** 8.3.2.44\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n> 11.[356]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=465.19.01\n\n> CUDA 11.3.0 GA | >=465.19.01 | >=465.89\n\n**Toolkit driver, Windows x86_64:** >=465.89\n\n> CUDA 11.3.0 GA | >=465.19.01 | >=465.89\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu113\n\n> | 1.12 | >=3.7, <=3.10 | C++14 | CUDA 11.3, CUDNN 8.3.2.44 | CUDA 11.6, CUDNN 8.3.2.44 | ROCm 5.0 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.3", "cudnn_version": "8.3.2.44", "file": "pytorch-1-12-with-cuda-11-3.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-1-12-with-cuda-11-3", "min_driver_family": ">= 450", "pairing": "PyTorch 1.12 with CUDA 11.3", "python_versions": ">=3.7, <=3.10", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "quote_arch_list_build": "11.[356]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.3.0 GA | >=465.19.01 | >=465.89", "quote_toolkit_driver_windows": "CUDA 11.3.0 GA | >=465.19.01 | >=465.89", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6", "source_quote": "| 1.12 | >=3.7, <=3.10 | C++14 | CUDA 11.3, CUDNN 8.3.2.44 | CUDA 11.6, CUDNN 8.3.2.44 | ROCm 5.0 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/1.12/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/1.12/windows/cuda113.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 1.12 with CUDA 11.3 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=465.19.01", "toolkit_driver_windows": ">=465.89", "torch_version": "1.12", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu113", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 1.12 with CUDA 11.6\n\n**PyTorch release:** 1.12\n\n**CUDA build:** CUDA 11.6\n\n**Python versions:** >=3.7, <=3.10\n\n**cuDNN version:** 8.3.2.44\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n> 11.[356]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=510.39.01\n\n> CUDA 11.6 GA | >=510.39.01 | >=511.23\n\n**Toolkit driver, Windows x86_64:** >=511.23\n\n> CUDA 11.6 GA | >=510.39.01 | >=511.23\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu116\n\n> | 1.12 | >=3.7, <=3.10 | C++14 | CUDA 11.3, CUDNN 8.3.2.44 | CUDA 11.6, CUDNN 8.3.2.44 | ROCm 5.0 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.6", "cudnn_version": "8.3.2.44", "file": "pytorch-1-12-with-cuda-11-6.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-1-12-with-cuda-11-6", "min_driver_family": ">= 450", "pairing": "PyTorch 1.12 with CUDA 11.6", "python_versions": ">=3.7, <=3.10", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "quote_arch_list_build": "11.[356]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.6 GA | >=510.39.01 | >=511.23", "quote_toolkit_driver_windows": "CUDA 11.6 GA | >=510.39.01 | >=511.23", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6", "source_quote": "| 1.12 | >=3.7, <=3.10 | C++14 | CUDA 11.3, CUDNN 8.3.2.44 | CUDA 11.6, CUDNN 8.3.2.44 | ROCm 5.0 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/1.12/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/1.12/windows/cuda116.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 1.12 with CUDA 11.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=510.39.01", "toolkit_driver_windows": ">=511.23", "torch_version": "1.12", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu116", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 1.13 with CUDA 11.6\n\n**PyTorch release:** 1.13\n\n**CUDA build:** CUDA 11.6\n\n**Python versions:** >=3.7, <=3.10\n\n**cuDNN version:** 8.3.2.44\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n> 11.[3567]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=510.39.01\n\n> CUDA 11.6 GA | >=510.39.01 | >=511.23\n\n**Toolkit driver, Windows x86_64:** >=511.23\n\n> CUDA 11.6 GA | >=510.39.01 | >=511.23\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu116\n\n> | 1.13 | >=3.7, <=3.10 | C++14 | CUDA 11.6, CUDNN 8.3.2.44 | CUDA 11.7, CUDNN 8.5.0.96 | ROCm 5.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.6", "cudnn_version": "8.3.2.44", "file": "pytorch-1-13-with-cuda-11-6.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-1-13-with-cuda-11-6", "min_driver_family": ">= 450", "pairing": "PyTorch 1.13 with CUDA 11.6", "python_versions": ">=3.7, <=3.10", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "quote_arch_list_build": "11.[3567]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.6 GA | >=510.39.01 | >=511.23", "quote_toolkit_driver_windows": "CUDA 11.6 GA | >=510.39.01 | >=511.23", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6", "source_quote": "| 1.13 | >=3.7, <=3.10 | C++14 | CUDA 11.6, CUDNN 8.3.2.44 | CUDA 11.7, CUDNN 8.5.0.96 | ROCm 5.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/1.13/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/1.13/windows/cuda116.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 1.13 with CUDA 11.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=510.39.01", "toolkit_driver_windows": ">=511.23", "torch_version": "1.13", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu116", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 1.13 with CUDA 11.7\n\n**PyTorch release:** 1.13\n\n**CUDA build:** CUDA 11.7\n\n**Python versions:** >=3.7, <=3.10\n\n**cuDNN version:** 8.5.0.96\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n> 11.[3567]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=515.43.04\n\n> CUDA 11.7 GA | >=515.43.04 | >=516.01\n\n**Toolkit driver, Windows x86_64:** >=516.01\n\n> CUDA 11.7 GA | >=515.43.04 | >=516.01\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu117\n\n> | 1.13 | >=3.7, <=3.10 | C++14 | CUDA 11.6, CUDNN 8.3.2.44 | CUDA 11.7, CUDNN 8.5.0.96 | ROCm 5.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.7", "cudnn_version": "8.5.0.96", "file": "pytorch-1-13-with-cuda-11-7.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-1-13-with-cuda-11-7", "min_driver_family": ">= 450", "pairing": "PyTorch 1.13 with CUDA 11.7", "python_versions": ">=3.7, <=3.10", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "quote_arch_list_build": "11.[3567]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.7 GA | >=515.43.04 | >=516.01", "quote_toolkit_driver_windows": "CUDA 11.7 GA | >=515.43.04 | >=516.01", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6", "source_quote": "| 1.13 | >=3.7, <=3.10 | C++14 | CUDA 11.6, CUDNN 8.3.2.44 | CUDA 11.7, CUDNN 8.5.0.96 | ROCm 5.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/1.13/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/1.13/windows/cuda117.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 1.13 with CUDA 11.7 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=515.43.04", "toolkit_driver_windows": ">=516.01", "torch_version": "1.13", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu117", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.0 with CUDA 11.7\n\n**PyTorch release:** 2.0\n\n**CUDA build:** CUDA 11.7\n\n**Python versions:** >=3.8, <=3.11\n\n**cuDNN version:** 8.5.0.96\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n> 11.[67]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=515.43.04\n\n> CUDA 11.7 GA | >=515.43.04 | >=516.01\n\n**Toolkit driver, Windows x86_64:** >=516.01\n\n> CUDA 11.7 GA | >=515.43.04 | >=516.01\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu117\n\n> | 2.0 | >=3.8, <=3.11 | C++14 | CUDA 11.7, CUDNN 8.5.0.96 | CUDA 11.8, CUDNN 8.7.0.84 | ROCm 5.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.7", "cudnn_version": "8.5.0.96", "file": "pytorch-2-0-with-cuda-11-7.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-0-with-cuda-11-7", "min_driver_family": ">= 450", "pairing": "PyTorch 2.0 with CUDA 11.7", "python_versions": ">=3.8, <=3.11", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "quote_arch_list_build": "11.[67]) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.7 GA | >=515.43.04 | >=516.01", "quote_toolkit_driver_windows": "CUDA 11.7 GA | >=515.43.04 | >=516.01", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6", "source_quote": "| 2.0 | >=3.8, <=3.11 | C++14 | CUDA 11.7, CUDNN 8.5.0.96 | CUDA 11.8, CUDNN 8.7.0.84 | ROCm 5.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.0/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.0/windows/cuda117.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.0 with CUDA 11.7 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=515.43.04", "toolkit_driver_windows": ">=516.01", "torch_version": "2.0", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu117", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6;9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.0 with CUDA 11.8\n\n**PyTorch release:** 2.0\n\n**CUDA build:** CUDA 11.8\n\n**Python versions:** >=3.8, <=3.11\n\n**cuDNN version:** 8.7.0.84\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6;9.0\"\n\n> 11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=520.61.05\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Toolkit driver, Windows x86_64:** >=520.06\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu118\n\n> | 2.0 | >=3.8, <=3.11 | C++14 | CUDA 11.7, CUDNN 8.5.0.96 | CUDA 11.8, CUDNN 8.7.0.84 | ROCm 5.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.8", "cudnn_version": "8.7.0.84", "file": "pytorch-2-0-with-cuda-11-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-0-with-cuda-11-8", "min_driver_family": ">= 450", "pairing": "PyTorch 2.0 with CUDA 11.8", "python_versions": ">=3.8, <=3.11", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"3.7;5.0;6.0;7.0\"", "quote_arch_list_build": "11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6;9.0\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_toolkit_driver_windows": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.0 | >=3.8, <=3.11 | C++14 | CUDA 11.7, CUDNN 8.5.0.96 | CUDA 11.8, CUDNN 8.7.0.84 | ROCm 5.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.0/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.0/windows/cuda118.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.0 with CUDA 11.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=520.61.05", "toolkit_driver_windows": ">=520.06", "torch_version": "2.0", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu118", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.1 with CUDA 11.8\n\n**PyTorch release:** 2.1\n\n**CUDA build:** CUDA 11.8\n\n**Python versions:** >=3.8, <=3.11\n\n**cuDNN version:** 8.7.0.84\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n> 11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=520.61.05\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Toolkit driver, Windows x86_64:** >=520.06\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu118\n\n> | 2.1 | >=3.8, <=3.11 | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 5.6 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.8", "cudnn_version": "8.7.0.84", "file": "pytorch-2-1-with-cuda-11-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-1-with-cuda-11-8", "min_driver_family": ">= 450", "pairing": "PyTorch 2.1 with CUDA 11.8", "python_versions": ">=3.8, <=3.11", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_toolkit_driver_windows": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.1 | >=3.8, <=3.11 | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 5.6 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.1/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.1/windows/cuda118.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.1 with CUDA 11.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=520.61.05", "toolkit_driver_windows": ">=520.06", "torch_version": "2.1", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu118", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.1 with CUDA 12.1\n\n**PyTorch release:** 2.1\n\n**CUDA build:** CUDA 12.1\n\n**Python versions:** >=3.8, <=3.11\n\n**cuDNN version:** 8.9.2.26\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=530.30.02\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Toolkit driver, Windows x86_64:** >=531.14\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu121\n\n> | 2.1 | >=3.8, <=3.11 | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 5.6 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.1", "cudnn_version": "8.9.2.26", "file": "pytorch-2-1-with-cuda-12-1.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-1-with-cuda-12-1", "min_driver_family": ">= 525", "pairing": "PyTorch 2.1 with CUDA 12.1", "python_versions": ">=3.8, <=3.11", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_toolkit_driver_windows": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.1 | >=3.8, <=3.11 | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 5.6 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.1/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.1/windows/cuda121.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.1 with CUDA 12.1 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=530.30.02", "toolkit_driver_windows": ">=531.14", "torch_version": "2.1", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu121", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;${TORCH_CUDA_ARCH_LIST}\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.10 with CUDA 12.6\n\n**PyTorch release:** 2.10\n\n**CUDA build:** CUDA 12.6\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;${TORCH_CUDA_ARCH_LIST}\"\n\n> 12.6) TORCH_CUDA_ARCH_LIST=\"5.0;6.0;${TORCH_CUDA_ARCH_LIST}\" ;; # Only 12.6 includes Legacy Maxwell/Pascal that will be removed in future releases\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=560.28.03\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Toolkit driver, Windows x86_64:** >=560.76\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu126\n\n> | 2.10 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.27.5), CUDA 12.8 (CUDNN 9.10.2.21) (NCCL 2.27.5) | CUDA 13.0 (CUDNN 9.15.1.9) (NCCL 2.28.9) | ROCm 7.1 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.6", "cudnn_version": "9.10.2.21", "file": "pytorch-2-10-with-cuda-12-6.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-10-with-cuda-12-6", "min_driver_family": ">= 525", "pairing": "PyTorch 2.10 with CUDA 12.6", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"", "quote_arch_list_build": "12.6) TORCH_CUDA_ARCH_LIST=\"5.0;6.0;${TORCH_CUDA_ARCH_LIST}\" ;; # Only 12.6 includes Legacy Maxwell/Pascal that will be removed in future releases", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_toolkit_driver_windows": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.10 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.27.5), CUDA 12.8 (CUDNN 9.10.2.21) (NCCL 2.27.5) | CUDA 13.0 (CUDNN 9.15.1.9) (NCCL 2.28.9) | ROCm 7.1 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/pytorch/windows/cuda126.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.10 with CUDA 12.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=560.28.03", "toolkit_driver_windows": ">=560.76", "torch_version": "2.10", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu126", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};10.0;12.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.10 with CUDA 12.8\n\n**PyTorch release:** 2.10\n\n**CUDA build:** CUDA 12.8\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};10.0;12.0\"\n\n> 12.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};10.0;12.0\" ;; # +Hopper/Blackwell support\n\n**GPU architectures compiled in, Windows wheel:** 7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=570.26\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Toolkit driver, Windows x86_64:** >=570.65\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu128\n\n> | 2.10 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.27.5), CUDA 12.8 (CUDNN 9.10.2.21) (NCCL 2.27.5) | CUDA 13.0 (CUDNN 9.15.1.9) (NCCL 2.28.9) | ROCm 7.1 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.8", "cudnn_version": "9.10.2.21", "file": "pytorch-2-10-with-cuda-12-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-10-with-cuda-12-8", "min_driver_family": ">= 525", "pairing": "PyTorch 2.10 with CUDA 12.8", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"", "quote_arch_list_build": "12.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};10.0;12.0\" ;; # +Hopper/Blackwell support", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_toolkit_driver_windows": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.0;7.5;8.0;8.6;9.0;10.0;12.0", "source_quote": "| 2.10 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.27.5), CUDA 12.8 (CUDNN 9.10.2.21) (NCCL 2.27.5) | CUDA 13.0 (CUDNN 9.15.1.9) (NCCL 2.28.9) | ROCm 7.1 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/pytorch/windows/cuda128.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.10 with CUDA 12.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=570.26", "toolkit_driver_windows": ">=570.65", "torch_version": "2.10", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu128", "windows_gpu_architectures": "7.0;7.5;8.0;8.6;9.0;10.0;12.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0;$([[ \"$ARCH\" == \"aarch64\" ]] && echo \"11.0;\" || echo \"\")12.0+PTX\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.10 with CUDA 13.0\n\n**PyTorch release:** 2.10\n\n**CUDA build:** CUDA 13.0\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.15.1.9\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0;$([[ \"$ARCH\" == \"aarch64\" ]] && echo \"11.0;\" || echo \"\")12.0+PTX\"\n\n> 13.0) TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0;$([[ \"$ARCH\" == \"aarch64\" ]] && echo \"11.0;\" || echo \"\")12.0+PTX\"\n\n**GPU architectures compiled in, Windows wheel:** 7.5;8.0;8.6;9.0;10.0;12.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\n\n**Minimum driver (minor version compatibility):** >= 580\n\n> 13.x | >= 580 | N/A\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu130\n\n> | 2.10 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.27.5), CUDA 12.8 (CUDNN 9.10.2.21) (NCCL 2.27.5) | CUDA 13.0 (CUDNN 9.15.1.9) (NCCL 2.28.9) | ROCm 7.1 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 13.0", "cudnn_version": "9.15.1.9", "file": "pytorch-2-10-with-cuda-13-0.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-10-with-cuda-13-0", "min_driver_family": ">= 580", "pairing": "PyTorch 2.10 with CUDA 13.0", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0\"", "quote_arch_list_build": "13.0) TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0;$([[ \"$ARCH\" == \"aarch64\" ]] && echo \"11.0;\" || echo \"\")12.0+PTX\"", "quote_min_driver_family": "13.x | >= 580 | N/A", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0", "source_quote": "| 2.10 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.27.5), CUDA 12.8 (CUDNN 9.10.2.21) (NCCL 2.27.5) | CUDA 13.0 (CUDNN 9.15.1.9) (NCCL 2.28.9) | ROCm 7.1 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.10/.ci/pytorch/windows/cuda130.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.10 with CUDA 13.0 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "torch_version": "2.10", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu130", "windows_gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;${TORCH_CUDA_ARCH_LIST//10.0/}\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.11 with CUDA 12.6\n\n**PyTorch release:** 2.11\n\n**CUDA build:** CUDA 12.6\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;${TORCH_CUDA_ARCH_LIST//10.0/}\"\n\n> 12.6) TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;${TORCH_CUDA_ARCH_LIST//10.0/}\" ;; # Only 12.6 includes legacy Maxwell/Pascal/Volta, -Hopper support\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> if \"%CUDA_VER%\"==\"126\" ( set \"CUDA_DOTTED_VERSION=12.6\" set \"CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\"\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=560.28.03\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Toolkit driver, Windows x86_64:** >=560.76\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu126\n\n> | 2.11 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.28.9), CUDA 12.8 (CUDNN 9.17.1.4) (NCCL 2.28.9), CUDA 13.0 (CUDNN 9.17.1.4) (NCCL 2.28.9) | -- | ROCm 7.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.6", "cudnn_version": "9.10.2.21", "file": "pytorch-2-11-with-cuda-12-6.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-11-with-cuda-12-6", "min_driver_family": ">= 525", "pairing": "PyTorch 2.11 with CUDA 12.6", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"", "quote_arch_list_build": "12.6) TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;${TORCH_CUDA_ARCH_LIST//10.0/}\" ;; # Only 12.6 includes legacy Maxwell/Pascal/Volta, -Hopper support", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_toolkit_driver_windows": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_windows_gpu_architectures": "if \"%CUDA_VER%\"==\"126\" ( set \"CUDA_DOTTED_VERSION=12.6\" set \"CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\"", "source_quote": "| 2.11 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.28.9), CUDA 12.8 (CUDNN 9.17.1.4) (NCCL 2.28.9), CUDA 13.0 (CUDNN 9.17.1.4) (NCCL 2.28.9) | -- | ROCm 7.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.11/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.11/.ci/pytorch/windows/internal/cuda_config.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.11 with CUDA 12.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=560.28.03", "toolkit_driver_windows": ">=560.76", "torch_version": "2.11", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu126", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};12.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.11 with CUDA 12.8\n\n**PyTorch release:** 2.11\n\n**CUDA build:** CUDA 12.8\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.17.1.4\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};12.0\"\n\n> 12.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};12.0\" ;; # +Blackwell support\n\n**GPU architectures compiled in, Windows wheel:** 7.5;8.0;8.6;9.0;10.0;12.0\n\n> ) else if \"%CUDA_VER%\"==\"128\" ( set \"CUDA_DOTTED_VERSION=12.8\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=570.26\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Toolkit driver, Windows x86_64:** >=570.65\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu128\n\n> | 2.11 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.28.9), CUDA 12.8 (CUDNN 9.17.1.4) (NCCL 2.28.9), CUDA 13.0 (CUDNN 9.17.1.4) (NCCL 2.28.9) | -- | ROCm 7.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.8", "cudnn_version": "9.17.1.4", "file": "pytorch-2-11-with-cuda-12-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-11-with-cuda-12-8", "min_driver_family": ">= 525", "pairing": "PyTorch 2.11 with CUDA 12.8", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"", "quote_arch_list_build": "12.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};12.0\" ;; # +Blackwell support", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_toolkit_driver_windows": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_windows_gpu_architectures": ") else if \"%CUDA_VER%\"==\"128\" ( set \"CUDA_DOTTED_VERSION=12.8\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"", "source_quote": "| 2.11 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.28.9), CUDA 12.8 (CUDNN 9.17.1.4) (NCCL 2.28.9), CUDA 13.0 (CUDNN 9.17.1.4) (NCCL 2.28.9) | -- | ROCm 7.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.11/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.11/.ci/pytorch/windows/internal/cuda_config.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.11 with CUDA 12.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=570.26", "toolkit_driver_windows": ">=570.65", "torch_version": "2.11", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu128", "windows_gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};$([[ \"$ARCH\" == \"aarch64\" ]] && echo \"11.0;\" || echo \"\")12.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.11 with CUDA 13.0\n\n**PyTorch release:** 2.11\n\n**CUDA build:** CUDA 13.0\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.17.1.4\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"\n\n> TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};$([[ \"$ARCH\" == \"aarch64\" ]] && echo \"11.0;\" || echo \"\")12.0\"\n\n> 13.0) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};$([[ \"$ARCH\" == \"aarch64\" ]] && echo \"11.0;\" || echo \"\")12.0\"\n\n**GPU architectures compiled in, Windows wheel:** 7.5;8.0;8.6;9.0;10.0;12.0\n\n> ) else if \"%CUDA_VER%\"==\"130\" ( set \"CUDA_DOTTED_VERSION=13.0\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"\n\n**Minimum driver (minor version compatibility):** >= 580\n\n> 13.x | >= 580 | N/A\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu130\n\n> | 2.11 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.28.9), CUDA 12.8 (CUDNN 9.17.1.4) (NCCL 2.28.9), CUDA 13.0 (CUDNN 9.17.1.4) (NCCL 2.28.9) | -- | ROCm 7.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 13.0", "cudnn_version": "9.17.1.4", "file": "pytorch-2-11-with-cuda-13-0.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-11-with-cuda-13-0", "min_driver_family": ">= 580", "pairing": "PyTorch 2.11 with CUDA 13.0", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0\"", "quote_arch_list_build": "13.0) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};$([[ \"$ARCH\" == \"aarch64\" ]] && echo \"11.0;\" || echo \"\")12.0\"", "quote_min_driver_family": "13.x | >= 580 | N/A", "quote_windows_gpu_architectures": ") else if \"%CUDA_VER%\"==\"130\" ( set \"CUDA_DOTTED_VERSION=13.0\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"", "source_quote": "| 2.11 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.28.9), CUDA 12.8 (CUDNN 9.17.1.4) (NCCL 2.28.9), CUDA 13.0 (CUDNN 9.17.1.4) (NCCL 2.28.9) | -- | ROCm 7.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.11/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.11/.ci/pytorch/windows/internal/cuda_config.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.11 with CUDA 13.0 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "torch_version": "2.11", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu130", "windows_gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.12 with CUDA 12.6\n\n**PyTorch release:** 2.12\n\n**CUDA build:** CUDA 12.6\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**GPU architectures compiled in, Linux wheel:** Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)\n\n> | 12.6.3 | Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0) | |\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> if \"%CUDA_VER%\"==\"126\" ( set \"CUDA_DOTTED_VERSION=12.6\" set \"CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\"\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=560.28.03\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Toolkit driver, Windows x86_64:** >=560.76\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu126\n\n> | 2.12 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.6", "cudnn_version": "9.10.2.21", "file": "pytorch-2-12-with-cuda-12-6.md", "generated": true, "gpu_architectures": "Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)", "harvested": "2026-09-16", "id": "pytorch-2-12-with-cuda-12-6", "min_driver_family": ">= 525", "pairing": "PyTorch 2.12 with CUDA 12.6", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_gpu_architectures": "| 12.6.3 | Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0) | |", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_toolkit_driver_windows": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_windows_gpu_architectures": "if \"%CUDA_VER%\"==\"126\" ( set \"CUDA_DOTTED_VERSION=12.6\" set \"CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\"", "source_quote": "| 2.12 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.12/.ci/pytorch/windows/internal/cuda_config.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.12 with CUDA 12.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=560.28.03", "toolkit_driver_windows": ">=560.76", "torch_version": "2.12", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu126", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.12 with CUDA 13.0\n\n**PyTorch release:** 2.12\n\n**CUDA build:** CUDA 13.0\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.20.0.48\n\n**GPU architectures compiled in, Linux wheel:** Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)\n\n> | 13.0.2 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |\n\n**GPU architectures compiled in, Windows wheel:** 7.5;8.0;8.6;9.0;10.0;12.0\n\n> ) else if \"%CUDA_VER%\"==\"130\" ( set \"CUDA_DOTTED_VERSION=13.0\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"\n\n**Minimum driver (minor version compatibility):** >= 580\n\n> 13.x | >= 580 | N/A\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu130\n\n> | 2.12 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 13.0", "cudnn_version": "9.20.0.48", "file": "pytorch-2-12-with-cuda-13-0.md", "generated": true, "gpu_architectures": "Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)", "harvested": "2026-09-16", "id": "pytorch-2-12-with-cuda-13-0", "min_driver_family": ">= 580", "pairing": "PyTorch 2.12 with CUDA 13.0", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_gpu_architectures": "| 13.0.2 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |", "quote_min_driver_family": "13.x | >= 580 | N/A", "quote_windows_gpu_architectures": ") else if \"%CUDA_VER%\"==\"130\" ( set \"CUDA_DOTTED_VERSION=13.0\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"", "source_quote": "| 2.12 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.12/.ci/pytorch/windows/internal/cuda_config.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.12 with CUDA 13.0 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "torch_version": "2.12", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu130", "windows_gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.12 with CUDA 13.2\n\n**PyTorch release:** 2.12\n\n**CUDA build:** CUDA 13.2\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.20.0.48\n\n**GPU architectures compiled in, Linux wheel:** Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)\n\n> | 13.2.1 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |\n\n**GPU architectures compiled in, Windows wheel:** 7.5;8.0;8.6;9.0;10.0;12.0\n\n> ) else if \"%CUDA_VER%\"==\"132\" ( set \"CUDA_DOTTED_VERSION=13.2\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"\n\n**Minimum driver (minor version compatibility):** >= 580\n\n> 13.x | >= 580 | N/A\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu132\n\n> | 2.12 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 13.2", "cudnn_version": "9.20.0.48", "file": "pytorch-2-12-with-cuda-13-2.md", "generated": true, "gpu_architectures": "Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)", "harvested": "2026-09-16", "id": "pytorch-2-12-with-cuda-13-2", "min_driver_family": ">= 580", "pairing": "PyTorch 2.12 with CUDA 13.2", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_gpu_architectures": "| 13.2.1 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |", "quote_min_driver_family": "13.x | >= 580 | N/A", "quote_windows_gpu_architectures": ") else if \"%CUDA_VER%\"==\"132\" ( set \"CUDA_DOTTED_VERSION=13.2\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"", "source_quote": "| 2.12 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.12/.ci/pytorch/windows/internal/cuda_config.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.12 with CUDA 13.2 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "torch_version": "2.12", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu132", "windows_gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.13 with CUDA 12.6\n\n**PyTorch release:** 2.13\n\n**CUDA build:** CUDA 12.6\n\n**Python versions:** >=3.10, <=(3.15, 3.15t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**GPU architectures compiled in, Linux wheel:** Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)\n\n> | 12.6.3 | Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0) | |\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> if \"%CUDA_VER%\"==\"126\" ( set \"CUDA_DOTTED_VERSION=12.6\" set \"CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\"\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=560.28.03\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Toolkit driver, Windows x86_64:** >=560.76\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu126\n\n> | 2.13 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.6", "cudnn_version": "9.10.2.21", "file": "pytorch-2-13-with-cuda-12-6.md", "generated": true, "gpu_architectures": "Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)", "harvested": "2026-09-16", "id": "pytorch-2-13-with-cuda-12-6", "min_driver_family": ">= 525", "pairing": "PyTorch 2.13 with CUDA 12.6", "python_versions": ">=3.10, <=(3.15, 3.15t experimental)", "quote_gpu_architectures": "| 12.6.3 | Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0) | |", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_toolkit_driver_windows": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_windows_gpu_architectures": "if \"%CUDA_VER%\"==\"126\" ( set \"CUDA_DOTTED_VERSION=12.6\" set \"CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\"", "source_quote": "| 2.13 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.13/.ci/pytorch/windows/internal/cuda_config.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.13 with CUDA 12.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=560.28.03", "toolkit_driver_windows": ">=560.76", "torch_version": "2.13", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu126", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.13 with CUDA 13.0\n\n**PyTorch release:** 2.13\n\n**CUDA build:** CUDA 13.0\n\n**Python versions:** >=3.10, <=(3.15, 3.15t experimental)\n\n**cuDNN version:** 9.20.0.48\n\n**GPU architectures compiled in, Linux wheel:** Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)\n\n> | 13.0.2 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |\n\n**GPU architectures compiled in, Windows wheel:** 7.5;8.0;8.6;9.0;10.0;12.0\n\n> ) else if \"%CUDA_VER%\"==\"130\" ( set \"CUDA_DOTTED_VERSION=13.0\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"\n\n**Minimum driver (minor version compatibility):** >= 580\n\n> 13.x | >= 580 | N/A\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu130\n\n> | 2.13 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 13.0", "cudnn_version": "9.20.0.48", "file": "pytorch-2-13-with-cuda-13-0.md", "generated": true, "gpu_architectures": "Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)", "harvested": "2026-09-16", "id": "pytorch-2-13-with-cuda-13-0", "min_driver_family": ">= 580", "pairing": "PyTorch 2.13 with CUDA 13.0", "python_versions": ">=3.10, <=(3.15, 3.15t experimental)", "quote_gpu_architectures": "| 13.0.2 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |", "quote_min_driver_family": "13.x | >= 580 | N/A", "quote_windows_gpu_architectures": ") else if \"%CUDA_VER%\"==\"130\" ( set \"CUDA_DOTTED_VERSION=13.0\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"", "source_quote": "| 2.13 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.13/.ci/pytorch/windows/internal/cuda_config.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.13 with CUDA 13.0 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "torch_version": "2.13", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu130", "windows_gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.13 with CUDA 13.2\n\n**PyTorch release:** 2.13\n\n**CUDA build:** CUDA 13.2\n\n**Python versions:** >=3.10, <=(3.15, 3.15t experimental)\n\n**cuDNN version:** 9.20.0.48\n\n**GPU architectures compiled in, Linux wheel:** Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)\n\n> | 13.2.1 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |\n\n**GPU architectures compiled in, Windows wheel:** 7.5;8.0;8.6;9.0;10.0;12.0\n\n> ) else if \"%CUDA_VER%\"==\"132\" ( set \"CUDA_DOTTED_VERSION=13.2\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"\n\n**Minimum driver (minor version compatibility):** >= 580\n\n> 13.x | >= 580 | N/A\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu132\n\n> | 2.13 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 13.2", "cudnn_version": "9.20.0.48", "file": "pytorch-2-13-with-cuda-13-2.md", "generated": true, "gpu_architectures": "Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)", "harvested": "2026-09-16", "id": "pytorch-2-13-with-cuda-13-2", "min_driver_family": ">= 580", "pairing": "PyTorch 2.13 with CUDA 13.2", "python_versions": ">=3.10, <=(3.15, 3.15t experimental)", "quote_gpu_architectures": "| 13.2.1 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |", "quote_min_driver_family": "13.x | >= 580 | N/A", "quote_windows_gpu_architectures": ") else if \"%CUDA_VER%\"==\"132\" ( set \"CUDA_DOTTED_VERSION=13.2\" set \"CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\"", "source_quote": "| 2.13 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.20.0.48) (NCCL 2.29.7) | CUDA 13.2 (CUDNN 9.20.0.48) (NCCL 2.29.7) | ROCm 7.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.13/.ci/pytorch/windows/internal/cuda_config.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.13 with CUDA 13.2 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "torch_version": "2.13", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu132", "windows_gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.14 with CUDA 12.6\n\n**PyTorch release:** 2.14\n\n**CUDA build:** CUDA 12.6\n\n**Python versions:** >=3.10, <=(3.15, 3.15t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**GPU architectures compiled in, Linux wheel:** Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)\n\n> | 12.6.3 | Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0) | |\n\n**GPU architectures compiled in, Windows wheel:** Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)\n\n> | 12.6.3 | Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0) | |\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=560.28.03\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Toolkit driver, Windows x86_64:** >=560.76\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu126\n\n> | 2.14 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.24.0.43) (NCCL 2.30.7), CUDA 13.2 (CUDNN 9.24.0.43) (NCCL 2.30.7) | -- | ROCm 7.14 |\n\nSource: \n\nAlso cited: \n", "cuda_build": "CUDA 12.6", "cudnn_version": "9.10.2.21", "file": "pytorch-2-14-with-cuda-12-6.md", "generated": true, "gpu_architectures": "Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)", "harvested": "2026-09-16", "id": "pytorch-2-14-with-cuda-12-6", "min_driver_family": ">= 525", "pairing": "PyTorch 2.14 with CUDA 12.6", "python_versions": ">=3.10, <=(3.15, 3.15t experimental)", "quote_gpu_architectures": "| 12.6.3 | Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0) | |", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_toolkit_driver_windows": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_windows_gpu_architectures": "| 12.6.3 | Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0) | |", "source_quote": "| 2.14 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.24.0.43) (NCCL 2.30.7), CUDA 13.2 (CUDNN 9.24.0.43) (NCCL 2.30.7) | -- | ROCm 7.14 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.14 with CUDA 12.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=560.28.03", "toolkit_driver_windows": ">=560.76", "torch_version": "2.14", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu126", "windows_gpu_architectures": "Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.14 with CUDA 13.0\n\n**PyTorch release:** 2.14\n\n**CUDA build:** CUDA 13.0\n\n**Python versions:** >=3.10, <=(3.15, 3.15t experimental)\n\n**cuDNN version:** 9.24.0.43\n\n**GPU architectures compiled in:** Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)\n\n> | 13.0.2 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |\n\n**Minimum driver (minor version compatibility):** >= 580\n\n> 13.x | >= 580 | N/A\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu130\n\n> | 2.14 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.24.0.43) (NCCL 2.30.7), CUDA 13.2 (CUDNN 9.24.0.43) (NCCL 2.30.7) | -- | ROCm 7.14 |\n\nSource: \n\nAlso cited: \n", "cuda_build": "CUDA 13.0", "cudnn_version": "9.24.0.43", "file": "pytorch-2-14-with-cuda-13-0.md", "generated": true, "gpu_architectures": "Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)", "harvested": "2026-09-14", "id": "pytorch-2-14-with-cuda-13-0", "min_driver_family": ">= 580", "pairing": "PyTorch 2.14 with CUDA 13.0", "python_versions": ">=3.10, <=(3.15, 3.15t experimental)", "quote_gpu_architectures": "| 13.0.2 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |", "quote_min_driver_family": "13.x | >= 580 | N/A", "source_quote": "| 2.14 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.24.0.43) (NCCL 2.30.7), CUDA 13.2 (CUDNN 9.24.0.43) (NCCL 2.30.7) | -- | ROCm 7.14 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-13", "title": "PyTorch 2.14 with CUDA 13.0 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "torch_version": "2.14", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu130"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.14 with CUDA 13.2\n\n**PyTorch release:** 2.14\n\n**CUDA build:** CUDA 13.2\n\n**Python versions:** >=3.10, <=(3.15, 3.15t experimental)\n\n**cuDNN version:** 9.24.0.43\n\n**GPU architectures compiled in:** Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)\n\n> | 13.2.1 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |\n\n**Minimum driver (minor version compatibility):** >= 580\n\n> 13.x | >= 580 | N/A\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu132\n\n> | 2.14 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.24.0.43) (NCCL 2.30.7), CUDA 13.2 (CUDNN 9.24.0.43) (NCCL 2.30.7) | -- | ROCm 7.14 |\n\nSource: \n\nAlso cited: \n", "cuda_build": "CUDA 13.2", "cudnn_version": "9.24.0.43", "file": "pytorch-2-14-with-cuda-13-2.md", "generated": true, "gpu_architectures": "Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)", "harvested": "2026-09-14", "id": "pytorch-2-14-with-cuda-13-2", "min_driver_family": ">= 580", "pairing": "PyTorch 2.14 with CUDA 13.2", "python_versions": ">=3.10, <=(3.15, 3.15t experimental)", "quote_gpu_architectures": "| 13.2.1 | Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX) | +PTX available on linux builds only |", "quote_min_driver_family": "13.x | >= 580 | N/A", "source_quote": "| 2.14 | >=3.10, <=(3.15, 3.15t experimental) | C++20 | CUDA 12.6 (CUDNN 9.10.2.21) (NCCL 2.29.3), CUDA 13.0 (CUDNN 9.24.0.43) (NCCL 2.30.7), CUDA 13.2 (CUDNN 9.24.0.43) (NCCL 2.30.7) | -- | ROCm 7.14 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-13", "title": "PyTorch 2.14 with CUDA 13.2 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "torch_version": "2.14", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu132"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.2 with CUDA 11.8\n\n**PyTorch release:** 2.2\n\n**CUDA build:** CUDA 11.8\n\n**Python versions:** >=3.8, <=3.11, (3.12 experimental)\n\n**cuDNN version:** 8.7.0.84\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n> 11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=520.61.05\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Toolkit driver, Windows x86_64:** >=520.06\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu118\n\n> | 2.2 | >=3.8, <=3.11, (3.12 experimental) | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 5.7 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.8", "cudnn_version": "8.7.0.84", "file": "pytorch-2-2-with-cuda-11-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-2-with-cuda-11-8", "min_driver_family": ">= 450", "pairing": "PyTorch 2.2 with CUDA 11.8", "python_versions": ">=3.8, <=3.11, (3.12 experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_toolkit_driver_windows": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.2 | >=3.8, <=3.11, (3.12 experimental) | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 5.7 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.2/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.2/windows/cuda118.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.2 with CUDA 11.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=520.61.05", "toolkit_driver_windows": ">=520.06", "torch_version": "2.2", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu118", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.2 with CUDA 12.1\n\n**PyTorch release:** 2.2\n\n**CUDA build:** CUDA 12.1\n\n**Python versions:** >=3.8, <=3.11, (3.12 experimental)\n\n**cuDNN version:** 8.9.2.26\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=530.30.02\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Toolkit driver, Windows x86_64:** >=531.14\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu121\n\n> | 2.2 | >=3.8, <=3.11, (3.12 experimental) | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 5.7 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.1", "cudnn_version": "8.9.2.26", "file": "pytorch-2-2-with-cuda-12-1.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-2-with-cuda-12-1", "min_driver_family": ">= 525", "pairing": "PyTorch 2.2 with CUDA 12.1", "python_versions": ">=3.8, <=3.11, (3.12 experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_toolkit_driver_windows": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.2 | >=3.8, <=3.11, (3.12 experimental) | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 5.7 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.2/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.2/windows/cuda121.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.2 with CUDA 12.1 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=530.30.02", "toolkit_driver_windows": ">=531.14", "torch_version": "2.2", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu121", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.3 with CUDA 11.8\n\n**PyTorch release:** 2.3\n\n**CUDA build:** CUDA 11.8\n\n**Python versions:** >=3.8, <=3.11, (3.12 experimental)\n\n**cuDNN version:** 8.7.0.84\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n> 11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=520.61.05\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Toolkit driver, Windows x86_64:** >=520.06\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu118\n\n> | 2.3 | >=3.8, <=3.11, (3.12 experimental) | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 6.0 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.8", "cudnn_version": "8.7.0.84", "file": "pytorch-2-3-with-cuda-11-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-3-with-cuda-11-8", "min_driver_family": ">= 450", "pairing": "PyTorch 2.3 with CUDA 11.8", "python_versions": ">=3.8, <=3.11, (3.12 experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_toolkit_driver_windows": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.3 | >=3.8, <=3.11, (3.12 experimental) | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 6.0 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.3/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.3/windows/cuda118.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.3 with CUDA 11.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=520.61.05", "toolkit_driver_windows": ">=520.06", "torch_version": "2.3", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu118", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.3 with CUDA 12.1\n\n**PyTorch release:** 2.3\n\n**CUDA build:** CUDA 12.1\n\n**Python versions:** >=3.8, <=3.11, (3.12 experimental)\n\n**cuDNN version:** 8.9.2.26\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=530.30.02\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Toolkit driver, Windows x86_64:** >=531.14\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu121\n\n> | 2.3 | >=3.8, <=3.11, (3.12 experimental) | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 6.0 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.1", "cudnn_version": "8.9.2.26", "file": "pytorch-2-3-with-cuda-12-1.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-3-with-cuda-12-1", "min_driver_family": ">= 525", "pairing": "PyTorch 2.3 with CUDA 12.1", "python_versions": ">=3.8, <=3.11, (3.12 experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_toolkit_driver_windows": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.3 | >=3.8, <=3.11, (3.12 experimental) | C++17 | CUDA 11.8, CUDNN 8.7.0.84 | CUDA 12.1, CUDNN 8.9.2.26 | ROCm 6.0 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.3/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.3/windows/cuda121.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.3 with CUDA 12.1 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=530.30.02", "toolkit_driver_windows": ">=531.14", "torch_version": "2.3", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu121", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.4 with CUDA 11.8\n\n**PyTorch release:** 2.4\n\n**CUDA build:** CUDA 11.8\n\n**Python versions:** >=3.8, <=3.12\n\n**cuDNN version:** 9.1.0.70\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n> 11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=520.61.05\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Toolkit driver, Windows x86_64:** >=520.06\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu118\n\n> | 2.4 | >=3.8, <=3.12 | C++17 | CUDA 11.8, CUDA 12.1, CUDNN 9.1.0.70 | CUDA 12.4, CUDNN 9.1.0.70 | ROCm 6.1 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.8", "cudnn_version": "9.1.0.70", "file": "pytorch-2-4-with-cuda-11-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-4-with-cuda-11-8", "min_driver_family": ">= 450", "pairing": "PyTorch 2.4 with CUDA 11.8", "python_versions": ">=3.8, <=3.12", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_toolkit_driver_windows": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.4 | >=3.8, <=3.12 | C++17 | CUDA 11.8, CUDA 12.1, CUDNN 9.1.0.70 | CUDA 12.4, CUDNN 9.1.0.70 | ROCm 6.1 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/windows/cuda118.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.4 with CUDA 11.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=520.61.05", "toolkit_driver_windows": ">=520.06", "torch_version": "2.4", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu118", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.4 with CUDA 12.1\n\n**PyTorch release:** 2.4\n\n**CUDA build:** CUDA 12.1\n\n**Python versions:** >=3.8, <=3.12\n\n**cuDNN version:** 9.1.0.70\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=530.30.02\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Toolkit driver, Windows x86_64:** >=531.14\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu121\n\n> | 2.4 | >=3.8, <=3.12 | C++17 | CUDA 11.8, CUDA 12.1, CUDNN 9.1.0.70 | CUDA 12.4, CUDNN 9.1.0.70 | ROCm 6.1 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.1", "cudnn_version": "9.1.0.70", "file": "pytorch-2-4-with-cuda-12-1.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-4-with-cuda-12-1", "min_driver_family": ">= 525", "pairing": "PyTorch 2.4 with CUDA 12.1", "python_versions": ">=3.8, <=3.12", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_toolkit_driver_windows": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.4 | >=3.8, <=3.12 | C++17 | CUDA 11.8, CUDA 12.1, CUDNN 9.1.0.70 | CUDA 12.4, CUDNN 9.1.0.70 | ROCm 6.1 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/windows/cuda121.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.4 with CUDA 12.1 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=530.30.02", "toolkit_driver_windows": ">=531.14", "torch_version": "2.4", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu121", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.4 with CUDA 12.4\n\n**PyTorch release:** 2.4\n\n**CUDA build:** CUDA 12.4\n\n**Python versions:** >=3.8, <=3.12\n\n**cuDNN version:** 9.1.0.70\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.4) if [[ \"$GPU_ARCH_TYPE\" = \"cuda-aarch64\" ]]; then TORCH_CUDA_ARCH_LIST=\"9.0\" else TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\" fi\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=550.54.14\n\n> CUDA 12.4 GA | >=550.54.14 | >=551.61\n\n**Toolkit driver, Windows x86_64:** >=551.61\n\n> CUDA 12.4 GA | >=550.54.14 | >=551.61\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu124\n\n> | 2.4 | >=3.8, <=3.12 | C++17 | CUDA 11.8, CUDA 12.1, CUDNN 9.1.0.70 | CUDA 12.4, CUDNN 9.1.0.70 | ROCm 6.1 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.4", "cudnn_version": "9.1.0.70", "file": "pytorch-2-4-with-cuda-12-4.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-4-with-cuda-12-4", "min_driver_family": ">= 525", "pairing": "PyTorch 2.4 with CUDA 12.4", "python_versions": ">=3.8, <=3.12", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.4) if [[ \"$GPU_ARCH_TYPE\" = \"cuda-aarch64\" ]]; then TORCH_CUDA_ARCH_LIST=\"9.0\" else TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\" fi", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.4 GA | >=550.54.14 | >=551.61", "quote_toolkit_driver_windows": "CUDA 12.4 GA | >=550.54.14 | >=551.61", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.4 | >=3.8, <=3.12 | C++17 | CUDA 11.8, CUDA 12.1, CUDNN 9.1.0.70 | CUDA 12.4, CUDNN 9.1.0.70 | ROCm 6.1 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.4/windows/cuda124.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.4 with CUDA 12.4 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=550.54.14", "toolkit_driver_windows": ">=551.61", "torch_version": "2.4", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu124", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.5 with CUDA 11.8\n\n**PyTorch release:** 2.5\n\n**CUDA build:** CUDA 11.8\n\n**Python versions:** >=3.9, <=3.12, (3.13 experimental)\n\n**cuDNN version:** 9.1.0.70\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n> 11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=520.61.05\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Toolkit driver, Windows x86_64:** >=520.06\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu118\n\n> | 2.5 | >=3.9, <=3.12, (3.13 experimental) | C++17 | CUDA 11.8, CUDA 12.1, CUDA 12.4, CUDNN 9.1.0.70 | None | ROCm 6.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.8", "cudnn_version": "9.1.0.70", "file": "pytorch-2-5-with-cuda-11-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-5-with-cuda-11-8", "min_driver_family": ">= 450", "pairing": "PyTorch 2.5 with CUDA 11.8", "python_versions": ">=3.9, <=3.12, (3.13 experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_toolkit_driver_windows": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.5 | >=3.9, <=3.12, (3.13 experimental) | C++17 | CUDA 11.8, CUDA 12.1, CUDA 12.4, CUDNN 9.1.0.70 | None | ROCm 6.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/windows/cuda118.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.5 with CUDA 11.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=520.61.05", "toolkit_driver_windows": ">=520.06", "torch_version": "2.5", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu118", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.5 with CUDA 12.1\n\n**PyTorch release:** 2.5\n\n**CUDA build:** CUDA 12.1\n\n**Python versions:** >=3.9, <=3.12, (3.13 experimental)\n\n**cuDNN version:** 9.1.0.70\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=530.30.02\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Toolkit driver, Windows x86_64:** >=531.14\n\n> CUDA 12.1 GA | >=530.30.02 | >=531.14\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu121\n\n> | 2.5 | >=3.9, <=3.12, (3.13 experimental) | C++17 | CUDA 11.8, CUDA 12.1, CUDA 12.4, CUDNN 9.1.0.70 | None | ROCm 6.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.1", "cudnn_version": "9.1.0.70", "file": "pytorch-2-5-with-cuda-12-1.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-5-with-cuda-12-1", "min_driver_family": ">= 525", "pairing": "PyTorch 2.5 with CUDA 12.1", "python_versions": ">=3.9, <=3.12, (3.13 experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.1) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_toolkit_driver_windows": "CUDA 12.1 GA | >=530.30.02 | >=531.14", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.5 | >=3.9, <=3.12, (3.13 experimental) | C++17 | CUDA 11.8, CUDA 12.1, CUDA 12.4, CUDNN 9.1.0.70 | None | ROCm 6.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/windows/cuda121.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.5 with CUDA 12.1 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=530.30.02", "toolkit_driver_windows": ">=531.14", "torch_version": "2.5", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu121", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.5 with CUDA 12.4\n\n**PyTorch release:** 2.5\n\n**CUDA build:** CUDA 12.4\n\n**Python versions:** >=3.9, <=3.12, (3.13 experimental)\n\n**cuDNN version:** 9.1.0.70\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.4) if [[ \"$GPU_ARCH_TYPE\" = \"cuda-aarch64\" ]]; then TORCH_CUDA_ARCH_LIST=\"9.0\" else TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\" fi\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=550.54.14\n\n> CUDA 12.4 GA | >=550.54.14 | >=551.61\n\n**Toolkit driver, Windows x86_64:** >=551.61\n\n> CUDA 12.4 GA | >=550.54.14 | >=551.61\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu124\n\n> | 2.5 | >=3.9, <=3.12, (3.13 experimental) | C++17 | CUDA 11.8, CUDA 12.1, CUDA 12.4, CUDNN 9.1.0.70 | None | ROCm 6.2 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.4", "cudnn_version": "9.1.0.70", "file": "pytorch-2-5-with-cuda-12-4.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-5-with-cuda-12-4", "min_driver_family": ">= 525", "pairing": "PyTorch 2.5 with CUDA 12.4", "python_versions": ">=3.9, <=3.12, (3.13 experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.4) if [[ \"$GPU_ARCH_TYPE\" = \"cuda-aarch64\" ]]; then TORCH_CUDA_ARCH_LIST=\"9.0\" else TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\" fi", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.4 GA | >=550.54.14 | >=551.61", "quote_toolkit_driver_windows": "CUDA 12.4 GA | >=550.54.14 | >=551.61", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.5 | >=3.9, <=3.12, (3.13 experimental) | C++17 | CUDA 11.8, CUDA 12.1, CUDA 12.4, CUDNN 9.1.0.70 | None | ROCm 6.2 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/builder/release/2.5/windows/cuda124.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.5 with CUDA 12.4 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=550.54.14", "toolkit_driver_windows": ">=551.61", "torch_version": "2.5", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu124", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.6 with CUDA 11.8\n\n**PyTorch release:** 2.6\n\n**CUDA build:** CUDA 11.8\n\n**Python versions:** >=3.9, <=3.13, (3.13t experimental)\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n> 11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=520.61.05\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Toolkit driver, Windows x86_64:** >=520.06\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu118\n\n> | 2.6 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8, CUDA 12.4 (CUDNN 9.1.0.70) | CUDA 12.6 (CUDNN 9.5.1.17) | ROCm 6.2.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.8", "file": "pytorch-2-6-with-cuda-11-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-6-with-cuda-11-8", "min_driver_family": ">= 450", "pairing": "PyTorch 2.6 with CUDA 11.8", "python_versions": ">=3.9, <=3.13, (3.13t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_toolkit_driver_windows": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.6 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8, CUDA 12.4 (CUDNN 9.1.0.70) | CUDA 12.6 (CUDNN 9.5.1.17) | ROCm 6.2.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/pytorch/windows/cuda118.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.6 with CUDA 11.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=520.61.05", "toolkit_driver_windows": ">=520.06", "torch_version": "2.6", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu118", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.6 with CUDA 12.4\n\n**PyTorch release:** 2.6\n\n**CUDA build:** CUDA 12.4\n\n**Python versions:** >=3.9, <=3.13, (3.13t experimental)\n\n**cuDNN version:** 9.1.0.70\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.4) if [[ \"$GPU_ARCH_TYPE\" = \"cuda-aarch64\" ]]; then TORCH_CUDA_ARCH_LIST=\"9.0\" else TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\" fi\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=550.54.14\n\n> CUDA 12.4 GA | >=550.54.14 | >=551.61\n\n**Toolkit driver, Windows x86_64:** >=551.61\n\n> CUDA 12.4 GA | >=550.54.14 | >=551.61\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu124\n\n> | 2.6 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8, CUDA 12.4 (CUDNN 9.1.0.70) | CUDA 12.6 (CUDNN 9.5.1.17) | ROCm 6.2.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.4", "cudnn_version": "9.1.0.70", "file": "pytorch-2-6-with-cuda-12-4.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-6-with-cuda-12-4", "min_driver_family": ">= 525", "pairing": "PyTorch 2.6 with CUDA 12.4", "python_versions": ">=3.9, <=3.13, (3.13t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.4) if [[ \"$GPU_ARCH_TYPE\" = \"cuda-aarch64\" ]]; then TORCH_CUDA_ARCH_LIST=\"9.0\" else TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\" fi", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.4 GA | >=550.54.14 | >=551.61", "quote_toolkit_driver_windows": "CUDA 12.4 GA | >=550.54.14 | >=551.61", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.6 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8, CUDA 12.4 (CUDNN 9.1.0.70) | CUDA 12.6 (CUDNN 9.5.1.17) | ROCm 6.2.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/pytorch/windows/cuda124.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.6 with CUDA 12.4 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=550.54.14", "toolkit_driver_windows": ">=551.61", "torch_version": "2.6", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu124", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.6 with CUDA 12.6\n\n**PyTorch release:** 2.6\n\n**CUDA build:** CUDA 12.6\n\n**Python versions:** >=3.9, <=3.13, (3.13t experimental)\n\n**cuDNN version:** 9.5.1.17\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.6) if [[ \"$GPU_ARCH_TYPE\" = \"cuda-aarch64\" ]]; then TORCH_CUDA_ARCH_LIST=\"9.0\" else TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\" fi\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=560.28.03\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Toolkit driver, Windows x86_64:** >=560.76\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu126\n\n> | 2.6 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8, CUDA 12.4 (CUDNN 9.1.0.70) | CUDA 12.6 (CUDNN 9.5.1.17) | ROCm 6.2.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.6", "cudnn_version": "9.5.1.17", "file": "pytorch-2-6-with-cuda-12-6.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-6-with-cuda-12-6", "min_driver_family": ">= 525", "pairing": "PyTorch 2.6 with CUDA 12.6", "python_versions": ">=3.9, <=3.13, (3.13t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.6) if [[ \"$GPU_ARCH_TYPE\" = \"cuda-aarch64\" ]]; then TORCH_CUDA_ARCH_LIST=\"9.0\" else TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\" fi", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_toolkit_driver_windows": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.6 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8, CUDA 12.4 (CUDNN 9.1.0.70) | CUDA 12.6 (CUDNN 9.5.1.17) | ROCm 6.2.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.6/.ci/pytorch/windows/cuda126.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.6 with CUDA 12.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=560.28.03", "toolkit_driver_windows": ">=560.76", "torch_version": "2.6", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu126", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.7 with CUDA 11.8\n\n**PyTorch release:** 2.7\n\n**CUDA build:** CUDA 11.8\n\n**Python versions:** >=3.9, <=3.13, (3.13t experimental)\n\n**cuDNN version:** 9.1.0.70\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n> 11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 450\n\n> 11.x | >= 450 | < 525\n\n**Toolkit driver, Linux x86_64:** >=520.61.05\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Toolkit driver, Windows x86_64:** >=520.06\n\n> CUDA 11.8 GA | >=520.61.05 | >=520.06\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu118\n\n> | 2.7 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8 (CUDNN 9.1.0.70), CUDA 12.6 (CUDNN 9.5.1.17) | CUDA 12.8 (CUDNN 9.7.1.26) | ROCm 6.3 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 11.8", "cudnn_version": "9.1.0.70", "file": "pytorch-2-7-with-cuda-11-8.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-7-with-cuda-11-8", "min_driver_family": ">= 450", "pairing": "PyTorch 2.7 with CUDA 11.8", "python_versions": ">=3.9, <=3.13, (3.13t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "11.8) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};3.7;9.0\"", "quote_min_driver_family": "11.x | >= 450 | < 525", "quote_toolkit_driver_linux": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_toolkit_driver_windows": "CUDA 11.8 GA | >=520.61.05 | >=520.06", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.7 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8 (CUDNN 9.1.0.70), CUDA 12.6 (CUDNN 9.5.1.17) | CUDA 12.8 (CUDNN 9.7.1.26) | ROCm 6.3 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/pytorch/windows/cuda118.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.7 with CUDA 11.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=520.61.05", "toolkit_driver_windows": ">=520.06", "torch_version": "2.7", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu118", "windows_gpu_architectures": "3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "arch_list_build": "TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.7 with CUDA 12.6\n\n**PyTorch release:** 2.7\n\n**CUDA build:** CUDA 12.6\n\n**Python versions:** >=3.9, <=3.13, (3.13t experimental)\n\n**cuDNN version:** 9.5.1.17\n\n**Build script: base architecture list:** TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n> TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"\n\n**Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n> 12.6) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=560.28.03\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Toolkit driver, Windows x86_64:** >=560.76\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu126\n\n> | 2.7 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8 (CUDNN 9.1.0.70), CUDA 12.6 (CUDNN 9.5.1.17) | CUDA 12.8 (CUDNN 9.7.1.26) | ROCm 6.3 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.6", "cudnn_version": "9.5.1.17", "file": "pytorch-2-7-with-cuda-12-6.md", "generated": true, "harvested": "2026-09-16", "id": "pytorch-2-7-with-cuda-12-6", "min_driver_family": ">= 525", "pairing": "PyTorch 2.7 with CUDA 12.6", "python_versions": ">=3.9, <=3.13, (3.13t experimental)", "quote_arch_list_base": "TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6\"", "quote_arch_list_build": "12.6) TORCH_CUDA_ARCH_LIST=\"${TORCH_CUDA_ARCH_LIST};9.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_toolkit_driver_windows": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.7 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8 (CUDNN 9.1.0.70), CUDA 12.6 (CUDNN 9.5.1.17) | CUDA 12.8 (CUDNN 9.7.1.26) | ROCm 6.3 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/pytorch/windows/cuda126.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.7 with CUDA 12.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=560.28.03", "toolkit_driver_windows": ">=560.76", "torch_version": "2.7", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu126", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.7 with CUDA 12.8\n\n**PyTorch release:** 2.7\n\n**CUDA build:** CUDA 12.8\n\n**Python versions:** >=3.9, <=3.13, (3.13t experimental)\n\n**cuDNN version:** 9.7.1.26\n\n**GPU architectures compiled in, Linux wheel:** 7.5;8.0;8.6;9.0;10.0;12.0+PTX\n\n> 12.8) TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0;12.0+PTX\" #removing sm_50-sm_70 as these architectures are deprecated in CUDA 12.8 and will be removed in future releases\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=570.26\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Toolkit driver, Windows x86_64:** >=570.65\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu128\n\n> | 2.7 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8 (CUDNN 9.1.0.70), CUDA 12.6 (CUDNN 9.5.1.17) | CUDA 12.8 (CUDNN 9.7.1.26) | ROCm 6.3 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.8", "cudnn_version": "9.7.1.26", "file": "pytorch-2-7-with-cuda-12-8.md", "generated": true, "gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0+PTX", "harvested": "2026-09-16", "id": "pytorch-2-7-with-cuda-12-8", "min_driver_family": ">= 525", "pairing": "PyTorch 2.7 with CUDA 12.8", "python_versions": ">=3.9, <=3.13, (3.13t experimental)", "quote_gpu_architectures": "12.8) TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0;12.0+PTX\" #removing sm_50-sm_70 as these architectures are deprecated in CUDA 12.8 and will be removed in future releases", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_toolkit_driver_windows": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0", "source_quote": "| 2.7 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 11.8 (CUDNN 9.1.0.70), CUDA 12.6 (CUDNN 9.5.1.17) | CUDA 12.8 (CUDNN 9.7.1.26) | ROCm 6.3 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.7/.ci/pytorch/windows/cuda128.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.7 with CUDA 12.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=570.26", "toolkit_driver_windows": ">=570.65", "torch_version": "2.7", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu128", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.8 with CUDA 12.6\n\n**PyTorch release:** 2.8\n\n**CUDA build:** CUDA 12.6\n\n**Python versions:** >=3.9, <=3.13, (3.13t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**GPU architectures compiled in, Linux wheel:** 5.0;6.0;7.0;7.5;8.0;8.6;9.0\n\n> 12.6) TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=560.28.03\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Toolkit driver, Windows x86_64:** >=560.76\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu126\n\n> | 2.8 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 12.9 (CUDNN 9.10.2.21) | ROCm 6.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.6", "cudnn_version": "9.10.2.21", "file": "pytorch-2-8-with-cuda-12-6.md", "generated": true, "gpu_architectures": "5.0;6.0;7.0;7.5;8.0;8.6;9.0", "harvested": "2026-09-16", "id": "pytorch-2-8-with-cuda-12-6", "min_driver_family": ">= 525", "pairing": "PyTorch 2.8 with CUDA 12.6", "python_versions": ">=3.9, <=3.13, (3.13t experimental)", "quote_gpu_architectures": "12.6) TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6;9.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_toolkit_driver_windows": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.8 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 12.9 (CUDNN 9.10.2.21) | ROCm 6.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/pytorch/windows/cuda126.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.8 with CUDA 12.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=560.28.03", "toolkit_driver_windows": ">=560.76", "torch_version": "2.8", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu126", "windows_gpu_architectures": "6.1;7.0;7.5;8.0;8.6;9.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.8 with CUDA 12.8\n\n**PyTorch release:** 2.8\n\n**CUDA build:** CUDA 12.8\n\n**Python versions:** >=3.9, <=3.13, (3.13t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**GPU architectures compiled in, Linux wheel:** 7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n> 12.8) TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0;10.0;12.0\"\n\n**GPU architectures compiled in, Windows wheel:** 6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=570.26\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Toolkit driver, Windows x86_64:** >=570.65\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu128\n\n> | 2.8 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 12.9 (CUDNN 9.10.2.21) | ROCm 6.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.8", "cudnn_version": "9.10.2.21", "file": "pytorch-2-8-with-cuda-12-8.md", "generated": true, "gpu_architectures": "7.0;7.5;8.0;8.6;9.0;10.0;12.0", "harvested": "2026-09-16", "id": "pytorch-2-8-with-cuda-12-8", "min_driver_family": ">= 525", "pairing": "PyTorch 2.8 with CUDA 12.8", "python_versions": ">=3.9, <=3.13, (3.13t experimental)", "quote_gpu_architectures": "12.8) TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0;10.0;12.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_toolkit_driver_windows": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0", "source_quote": "| 2.8 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 12.9 (CUDNN 9.10.2.21) | ROCm 6.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/pytorch/windows/cuda128.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.8 with CUDA 12.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=570.26", "toolkit_driver_windows": ">=570.65", "torch_version": "2.8", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu128", "windows_gpu_architectures": "6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.8 with CUDA 12.9\n\n**PyTorch release:** 2.8\n\n**CUDA build:** CUDA 12.9\n\n**Python versions:** >=3.9, <=3.13, (3.13t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**GPU architectures compiled in, Linux wheel:** 7.0;7.5;8.0;8.6;9.0;10.0;12.0+PTX\n\n> 12.9) TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0;10.0;12.0+PTX\"\n\n**GPU architectures compiled in, Windows wheel:** 7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=575.51.03\n\n> CUDA 12.9 GA | >=575.51.03 | >=576.02\n\n**Toolkit driver, Windows x86_64:** >=576.02\n\n> CUDA 12.9 GA | >=575.51.03 | >=576.02\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu129\n\n> | 2.8 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 12.9 (CUDNN 9.10.2.21) | ROCm 6.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.9", "cudnn_version": "9.10.2.21", "file": "pytorch-2-8-with-cuda-12-9.md", "generated": true, "gpu_architectures": "7.0;7.5;8.0;8.6;9.0;10.0;12.0+PTX", "harvested": "2026-09-16", "id": "pytorch-2-8-with-cuda-12-9", "min_driver_family": ">= 525", "pairing": "PyTorch 2.8 with CUDA 12.9", "python_versions": ">=3.9, <=3.13, (3.13t experimental)", "quote_gpu_architectures": "12.9) TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0;10.0;12.0+PTX\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.9 GA | >=575.51.03 | >=576.02", "quote_toolkit_driver_windows": "CUDA 12.9 GA | >=575.51.03 | >=576.02", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.0;7.5;8.0;8.6;9.0;10.0;12.0", "source_quote": "| 2.8 | >=3.9, <=3.13, (3.13t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 12.9 (CUDNN 9.10.2.21) | ROCm 6.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.8/.ci/pytorch/windows/cuda129.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.8 with CUDA 12.9 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=575.51.03", "toolkit_driver_windows": ">=576.02", "torch_version": "2.8", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu129", "windows_gpu_architectures": "7.0;7.5;8.0;8.6;9.0;10.0;12.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.9 with CUDA 12.6\n\n**PyTorch release:** 2.9\n\n**CUDA build:** CUDA 12.6\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**GPU architectures compiled in, Linux wheel:** 5.0;6.0;7.0;7.5;8.0;8.6;9.0\n\n> 12.6) TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6;9.0\"\n\n**GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=560.28.03\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Toolkit driver, Windows x86_64:** >=560.76\n\n> CUDA 12.6 GA | >=560.28.03 | >=560.76\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu126\n\n> | 2.9 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 13.0 (CUDNN 9.13.0.50) | ROCm 6.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.6", "cudnn_version": "9.10.2.21", "file": "pytorch-2-9-with-cuda-12-6.md", "generated": true, "gpu_architectures": "5.0;6.0;7.0;7.5;8.0;8.6;9.0", "harvested": "2026-09-16", "id": "pytorch-2-9-with-cuda-12-6", "min_driver_family": ">= 525", "pairing": "PyTorch 2.9 with CUDA 12.6", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_gpu_architectures": "12.6) TORCH_CUDA_ARCH_LIST=\"5.0;6.0;7.0;7.5;8.0;8.6;9.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_toolkit_driver_windows": "CUDA 12.6 GA | >=560.28.03 | >=560.76", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0", "source_quote": "| 2.9 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 13.0 (CUDNN 9.13.0.50) | ROCm 6.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/pytorch/windows/cuda126.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.9 with CUDA 12.6 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=560.28.03", "toolkit_driver_windows": ">=560.76", "torch_version": "2.9", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu126", "windows_gpu_architectures": "5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.9 with CUDA 12.8\n\n**PyTorch release:** 2.9\n\n**CUDA build:** CUDA 12.8\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.10.2.21\n\n**GPU architectures compiled in, Linux wheel:** 7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n> 12.8) TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0;10.0;12.0\"\n\n**GPU architectures compiled in, Windows wheel:** 7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.0;7.5;8.0;8.6;9.0;10.0;12.0\n\n**Minimum driver (minor version compatibility):** >= 525\n\n> 12.x | >= 525 | < 580\n\n**Toolkit driver, Linux x86_64:** >=570.26\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Toolkit driver, Windows x86_64:** >=570.65\n\n> CUDA 12.8 GA | >=570.26 | >=570.65\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu128\n\n> | 2.9 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 13.0 (CUDNN 9.13.0.50) | ROCm 6.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 12.8", "cudnn_version": "9.10.2.21", "file": "pytorch-2-9-with-cuda-12-8.md", "generated": true, "gpu_architectures": "7.0;7.5;8.0;8.6;9.0;10.0;12.0", "harvested": "2026-09-16", "id": "pytorch-2-9-with-cuda-12-8", "min_driver_family": ">= 525", "pairing": "PyTorch 2.9 with CUDA 12.8", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_gpu_architectures": "12.8) TORCH_CUDA_ARCH_LIST=\"7.0;7.5;8.0;8.6;9.0;10.0;12.0\"", "quote_min_driver_family": "12.x | >= 525 | < 580", "quote_toolkit_driver_linux": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_toolkit_driver_windows": "CUDA 12.8 GA | >=570.26 | >=570.65", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.0;7.5;8.0;8.6;9.0;10.0;12.0", "source_quote": "| 2.9 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 13.0 (CUDNN 9.13.0.50) | ROCm 6.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/pytorch/windows/cuda128.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.9 with CUDA 12.8 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "toolkit_driver_linux": ">=570.26", "toolkit_driver_windows": ">=570.65", "torch_version": "2.9", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu128", "windows_gpu_architectures": "7.0;7.5;8.0;8.6;9.0;10.0;12.0"} {"asset": "gpu-cuda-pytorch-compatibility", "body": "**Pairing:** PyTorch 2.9 with CUDA 13.0\n\n**PyTorch release:** 2.9\n\n**CUDA build:** CUDA 13.0\n\n**Python versions:** >=3.10, <=(3.14, 3.14t experimental)\n\n**cuDNN version:** 9.13.0.50\n\n**GPU architectures compiled in, Linux wheel:** 7.5;8.0;8.6;9.0;10.0;12.0+PTX\n\n> 13.0) TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0;12.0+PTX\"\n\n**GPU architectures compiled in, Windows wheel:** 7.5;8.0;8.6;9.0;10.0;12.0\n\n> IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0\n\n**Minimum driver (minor version compatibility):** >= 580\n\n> 13.x | >= 580 | N/A\n\n**Wheel index URL:** https://download.pytorch.org/whl/cu130\n\n> | 2.9 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 13.0 (CUDNN 9.13.0.50) | ROCm 6.4 |\n\nSource: \n\nAlso cited: \n\nAlso cited: \n\nAlso cited: \n", "cuda_build": "CUDA 13.0", "cudnn_version": "9.13.0.50", "file": "pytorch-2-9-with-cuda-13-0.md", "generated": true, "gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0+PTX", "harvested": "2026-09-16", "id": "pytorch-2-9-with-cuda-13-0", "min_driver_family": ">= 580", "pairing": "PyTorch 2.9 with CUDA 13.0", "python_versions": ">=3.10, <=(3.14, 3.14t experimental)", "quote_gpu_architectures": "13.0) TORCH_CUDA_ARCH_LIST=\"7.5;8.0;8.6;9.0;10.0;12.0+PTX\"", "quote_min_driver_family": "13.x | >= 580 | N/A", "quote_windows_gpu_architectures": "IF \"%BUILD_VISION%\" == \"\" ( set TORCH_CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0", "source_quote": "| 2.9 | >=3.10, <=(3.14, 3.14t experimental) | C++17 | CUDA 12.6 (CUDNN 9.10.2.21), CUDA 12.8 (CUDNN 9.10.2.21) | CUDA 13.0 (CUDNN 9.13.0.50) | ROCm 6.4 |", "sources": ["https://raw.githubusercontent.com/pytorch/pytorch/main/RELEASE.md", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/manywheel/build_cuda.sh", "https://raw.githubusercontent.com/pytorch/pytorch/release/2.9/.ci/pytorch/windows/cuda130.bat", "https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html"], "stale_after": "2026-12-15", "title": "PyTorch 2.9 with CUDA 13.0 \u2014 PyTorch release, CUDA build and NVIDIA driver pairings", "torch_version": "2.9", "type": "lookup", "verified": false, "wheel_index_url": "https://download.pytorch.org/whl/cu130", "windows_gpu_architectures": "7.5;8.0;8.6;9.0;10.0;12.0"}