# PyTorch 2.5 with CUDA 12.4 — PyTorch release, CUDA build and NVIDIA driver pairings

For PyTorch 2.5 with CUDA 12.4, pytorch release is 2.5; cuda build is CUDA 12.4; python versions is >=3.9, <=3.12, (3.13 experimental); cudnn version is 9.1.0.70; build script: base architecture list is TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6", recorded from its source on 2026-09-16.

- **Pairing:** PyTorch 2.5 with CUDA 12.4 _(our reading, not quoted from the source)_
- **PyTorch release:** 2.5 _(verified: appears in the quote below)_
- **CUDA build:** CUDA 12.4 _(verified: appears in the quote below)_
- **Python versions:** >=3.9, <=3.12, (3.13 experimental) _(verified: appears in the quote below)_
- **cuDNN version:** 9.1.0.70 _(verified: appears in the quote below)_
- **Build script: base architecture list:** TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6" _(verified: appears in its own passage below)_
- **Build script: this CUDA build's architecture line:** TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0" _(verified: appears in its own passage below)_
- **GPU architectures compiled in, Windows wheel:** 5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0 _(verified: appears in its own passage below)_
- **Minimum driver (minor version compatibility):** >= 525 _(verified: appears in its own passage below)_
- **Toolkit driver, Linux x86_64:** >=550.54.14 _(verified: appears in its own passage below)_
- **Toolkit driver, Windows x86_64:** >=551.61 _(verified: appears in its own passage below)_
- **Wheel index URL:** https://download.pytorch.org/whl/cu124 _(our reading, not quoted from the source)_

## What the source says

> | 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 |

## Where each value comes from

This source states these in separate places, so each value is shown with the passage that states it.

### Build script: base architecture list

> TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"

### Build script: this CUDA build's architecture line

> 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

### GPU architectures compiled in, Windows wheel

> IF "%BUILD_VISION%" == "" ( set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0

### Minimum driver (minor version compatibility)

> 12.x | >= 525 | < 580

### Toolkit driver, Linux x86_64

> CUDA 12.4 GA | >=550.54.14 | >=551.61

### Toolkit driver, Windows x86_64

> CUDA 12.4 GA | >=550.54.14 | >=551.61

## Source

- 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

Last verified: 2026-09-16. Review by: 2026-12-15.
Part of [PyTorch release, CUDA build and NVIDIA driver pairings](https://referencesource.org/gpu-cuda-pytorch-compatibility/).