Reference Source

PyTorch release, CUDA build and NVIDIA driver pairings

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 — 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.

Records45
Sources52
Verified
Review by
LicencePyTorch'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.

The data

PairingPyTorch releaseCUDA buildPython versionscuDNN versionBuild script: base architecture listBuild script: this CUDA build's architecture lineGPU architectures compiled in, Windows wheelMinimum driver (minor version compatibility)Toolkit driver, Linux x86_64Toolkit driver, Windows x86_64Wheel index URLGPU architectures compiled in, Linux wheel
PyTorch 1.12 with CUDA 11.31.12CUDA 11.3>=3.7, <=3.108.3.2.44TORCH_CUDA_ARCH_LIST="3.7;5.0;6.0;7.0"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6>= 450>=465.19.01>=465.89https://download.pytorch.org/whl/cu113
PyTorch 1.12 with CUDA 11.61.12CUDA 11.6>=3.7, <=3.108.3.2.44TORCH_CUDA_ARCH_LIST="3.7;5.0;6.0;7.0"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6>= 450>=510.39.01>=511.23https://download.pytorch.org/whl/cu116
PyTorch 1.13 with CUDA 11.61.13CUDA 11.6>=3.7, <=3.108.3.2.44TORCH_CUDA_ARCH_LIST="3.7;5.0;6.0;7.0"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6>= 450>=510.39.01>=511.23https://download.pytorch.org/whl/cu116
PyTorch 1.13 with CUDA 11.71.13CUDA 11.7>=3.7, <=3.108.5.0.96TORCH_CUDA_ARCH_LIST="3.7;5.0;6.0;7.0"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6>= 450>=515.43.04>=516.01https://download.pytorch.org/whl/cu117
PyTorch 2.0 with CUDA 11.72.0CUDA 11.7>=3.8, <=3.118.5.0.96TORCH_CUDA_ARCH_LIST="3.7;5.0;6.0;7.0"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6>= 450>=515.43.04>=516.01https://download.pytorch.org/whl/cu117
PyTorch 2.0 with CUDA 11.82.0CUDA 11.8>=3.8, <=3.118.7.0.84TORCH_CUDA_ARCH_LIST="3.7;5.0;6.0;7.0"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6;9.0"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 450>=520.61.05>=520.06https://download.pytorch.org/whl/cu118
PyTorch 2.1 with CUDA 11.82.1CUDA 11.8>=3.8, <=3.118.7.0.84TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 450>=520.61.05>=520.06https://download.pytorch.org/whl/cu118
PyTorch 2.1 with CUDA 12.12.1CUDA 12.1>=3.8, <=3.118.9.2.26TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=530.30.02>=531.14https://download.pytorch.org/whl/cu121
PyTorch 2.10 with CUDA 12.62.10CUDA 12.6>=3.10, <=(3.14, 3.14t experimental)9.10.2.21TORCH_CUDA_ARCH_LIST="7.0;7.5;8.0;8.6;9.0"TORCH_CUDA_ARCH_LIST="5.0;6.0;${TORCH_CUDA_ARCH_LIST}"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=560.28.03>=560.76https://download.pytorch.org/whl/cu126
PyTorch 2.10 with CUDA 12.82.10CUDA 12.8>=3.10, <=(3.14, 3.14t experimental)9.10.2.21TORCH_CUDA_ARCH_LIST="7.0;7.5;8.0;8.6;9.0"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};10.0;12.0"7.0;7.5;8.0;8.6;9.0;10.0;12.0>= 525>=570.26>=570.65https://download.pytorch.org/whl/cu128
PyTorch 2.10 with CUDA 13.02.10CUDA 13.0>=3.10, <=(3.14, 3.14t experimental)9.15.1.9TORCH_CUDA_ARCH_LIST="7.0;7.5;8.0;8.6;9.0"TORCH_CUDA_ARCH_LIST="7.5;8.0;8.6;9.0;10.0;$([[ "$ARCH" == "aarch64" ]] && echo "11.0;" || echo "")12.0+PTX"7.5;8.0;8.6;9.0;10.0;12.0>= 580https://download.pytorch.org/whl/cu130
PyTorch 2.11 with CUDA 12.62.11CUDA 12.6>=3.10, <=(3.14, 3.14t experimental)9.10.2.21TORCH_CUDA_ARCH_LIST="7.5;8.0;8.6;9.0;10.0"TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;${TORCH_CUDA_ARCH_LIST//10.0/}"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=560.28.03>=560.76https://download.pytorch.org/whl/cu126
PyTorch 2.11 with CUDA 12.82.11CUDA 12.8>=3.10, <=(3.14, 3.14t experimental)9.17.1.4TORCH_CUDA_ARCH_LIST="7.5;8.0;8.6;9.0;10.0"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};12.0"7.5;8.0;8.6;9.0;10.0;12.0>= 525>=570.26>=570.65https://download.pytorch.org/whl/cu128
PyTorch 2.11 with CUDA 13.02.11CUDA 13.0>=3.10, <=(3.14, 3.14t experimental)9.17.1.4TORCH_CUDA_ARCH_LIST="7.5;8.0;8.6;9.0;10.0"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};$([[ "$ARCH" == "aarch64" ]] && echo "11.0;" || echo "")12.0"7.5;8.0;8.6;9.0;10.0;12.0>= 580https://download.pytorch.org/whl/cu130
PyTorch 2.12 with CUDA 12.62.12CUDA 12.6>=3.10, <=(3.14, 3.14t experimental)9.10.2.215.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=560.28.03>=560.76https://download.pytorch.org/whl/cu126Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)
PyTorch 2.12 with CUDA 13.02.12CUDA 13.0>=3.10, <=(3.14, 3.14t experimental)9.20.0.487.5;8.0;8.6;9.0;10.0;12.0>= 580https://download.pytorch.org/whl/cu130Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)
PyTorch 2.12 with CUDA 13.22.12CUDA 13.2>=3.10, <=(3.14, 3.14t experimental)9.20.0.487.5;8.0;8.6;9.0;10.0;12.0>= 580https://download.pytorch.org/whl/cu132Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)
PyTorch 2.13 with CUDA 12.62.13CUDA 12.6>=3.10, <=(3.15, 3.15t experimental)9.10.2.215.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=560.28.03>=560.76https://download.pytorch.org/whl/cu126Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)
PyTorch 2.13 with CUDA 13.02.13CUDA 13.0>=3.10, <=(3.15, 3.15t experimental)9.20.0.487.5;8.0;8.6;9.0;10.0;12.0>= 580https://download.pytorch.org/whl/cu130Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)
PyTorch 2.13 with CUDA 13.22.13CUDA 13.2>=3.10, <=(3.15, 3.15t experimental)9.20.0.487.5;8.0;8.6;9.0;10.0;12.0>= 580https://download.pytorch.org/whl/cu132Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)
PyTorch 2.14 with CUDA 12.62.14CUDA 12.6>=3.10, <=(3.15, 3.15t experimental)9.10.2.21Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)>= 525>=560.28.03>=560.76https://download.pytorch.org/whl/cu126Maxwell(5.0), Pascal(6.0), Volta(7.0), Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0)
PyTorch 2.14 with CUDA 13.02.14CUDA 13.0>=3.10, <=(3.15, 3.15t experimental)9.24.0.43>= 580https://download.pytorch.org/whl/cu130Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)
PyTorch 2.14 with CUDA 13.22.14CUDA 13.2>=3.10, <=(3.15, 3.15t experimental)9.24.0.43>= 580https://download.pytorch.org/whl/cu132Turing(7.5), Ampere(8.0, 8.6), Hopper(9.0), Blackwell(10.0, 12.0+PTX)
PyTorch 2.2 with CUDA 11.82.2CUDA 11.8>=3.8, <=3.11, (3.12 experimental)8.7.0.84TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 450>=520.61.05>=520.06https://download.pytorch.org/whl/cu118
PyTorch 2.2 with CUDA 12.12.2CUDA 12.1>=3.8, <=3.11, (3.12 experimental)8.9.2.26TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=530.30.02>=531.14https://download.pytorch.org/whl/cu121
PyTorch 2.3 with CUDA 11.82.3CUDA 11.8>=3.8, <=3.11, (3.12 experimental)8.7.0.84TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 450>=520.61.05>=520.06https://download.pytorch.org/whl/cu118
PyTorch 2.3 with CUDA 12.12.3CUDA 12.1>=3.8, <=3.11, (3.12 experimental)8.9.2.26TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=530.30.02>=531.14https://download.pytorch.org/whl/cu121
PyTorch 2.4 with CUDA 11.82.4CUDA 11.8>=3.8, <=3.129.1.0.70TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 450>=520.61.05>=520.06https://download.pytorch.org/whl/cu118
PyTorch 2.4 with CUDA 12.12.4CUDA 12.1>=3.8, <=3.129.1.0.70TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=530.30.02>=531.14https://download.pytorch.org/whl/cu121
PyTorch 2.4 with CUDA 12.42.4CUDA 12.4>=3.8, <=3.129.1.0.70TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=550.54.14>=551.61https://download.pytorch.org/whl/cu124
PyTorch 2.5 with CUDA 11.82.5CUDA 11.8>=3.9, <=3.12, (3.13 experimental)9.1.0.70TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 450>=520.61.05>=520.06https://download.pytorch.org/whl/cu118
PyTorch 2.5 with CUDA 12.12.5CUDA 12.1>=3.9, <=3.12, (3.13 experimental)9.1.0.70TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=530.30.02>=531.14https://download.pytorch.org/whl/cu121
PyTorch 2.5 with CUDA 12.42.5CUDA 12.4>=3.9, <=3.12, (3.13 experimental)9.1.0.70TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=550.54.14>=551.61https://download.pytorch.org/whl/cu124
PyTorch 2.6 with CUDA 11.82.6CUDA 11.8>=3.9, <=3.13, (3.13t experimental)TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 450>=520.61.05>=520.06https://download.pytorch.org/whl/cu118
PyTorch 2.6 with CUDA 12.42.6CUDA 12.4>=3.9, <=3.13, (3.13t experimental)9.1.0.70TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=550.54.14>=551.61https://download.pytorch.org/whl/cu124
PyTorch 2.6 with CUDA 12.62.6CUDA 12.6>=3.9, <=3.13, (3.13t experimental)9.5.1.17TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=560.28.03>=560.76https://download.pytorch.org/whl/cu126
PyTorch 2.7 with CUDA 11.82.7CUDA 11.8>=3.9, <=3.13, (3.13t experimental)9.1.0.70TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 450>=520.61.05>=520.06https://download.pytorch.org/whl/cu118
PyTorch 2.7 with CUDA 12.62.7CUDA 12.6>=3.9, <=3.13, (3.13t experimental)9.5.1.17TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=560.28.03>=560.76https://download.pytorch.org/whl/cu126
PyTorch 2.7 with CUDA 12.82.7CUDA 12.8>=3.9, <=3.13, (3.13t experimental)9.7.1.265.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0>= 525>=570.26>=570.65https://download.pytorch.org/whl/cu1287.5;8.0;8.6;9.0;10.0;12.0+PTX
PyTorch 2.8 with CUDA 12.62.8CUDA 12.6>=3.9, <=3.13, (3.13t experimental)9.10.2.216.1;7.0;7.5;8.0;8.6;9.0>= 525>=560.28.03>=560.76https://download.pytorch.org/whl/cu1265.0;6.0;7.0;7.5;8.0;8.6;9.0
PyTorch 2.8 with CUDA 12.82.8CUDA 12.8>=3.9, <=3.13, (3.13t experimental)9.10.2.216.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0>= 525>=570.26>=570.65https://download.pytorch.org/whl/cu1287.0;7.5;8.0;8.6;9.0;10.0;12.0
PyTorch 2.8 with CUDA 12.92.8CUDA 12.9>=3.9, <=3.13, (3.13t experimental)9.10.2.217.0;7.5;8.0;8.6;9.0;10.0;12.0>= 525>=575.51.03>=576.02https://download.pytorch.org/whl/cu1297.0;7.5;8.0;8.6;9.0;10.0;12.0+PTX
PyTorch 2.9 with CUDA 12.62.9CUDA 12.6>=3.10, <=(3.14, 3.14t experimental)9.10.2.215.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0>= 525>=560.28.03>=560.76https://download.pytorch.org/whl/cu1265.0;6.0;7.0;7.5;8.0;8.6;9.0
PyTorch 2.9 with CUDA 12.82.9CUDA 12.8>=3.10, <=(3.14, 3.14t experimental)9.10.2.217.0;7.5;8.0;8.6;9.0;10.0;12.0>= 525>=570.26>=570.65https://download.pytorch.org/whl/cu1287.0;7.5;8.0;8.6;9.0;10.0;12.0
PyTorch 2.9 with CUDA 13.02.9CUDA 13.0>=3.10, <=(3.14, 3.14t experimental)9.13.0.507.5;8.0;8.6;9.0;10.0;12.0>= 580https://download.pytorch.org/whl/cu1307.5;8.0;8.6;9.0;10.0;12.0+PTX

Where this came from

Every record above links the page it was taken from and quotes the sentence that states it. These are the 52 sources this dataset was assembled from.

Machine-readable

Set your AI assistant up to use this

Two files and no account. Put this in .mcp.json at the root of your project — Claude Code, Cursor, Windsurf, VS Code and Codex all read that file — and your assistant can look this dataset up instead of recalling it. The server is remote, keyless and read-only.

{
  "mcpServers": {
    "referencesource": {
      "type": "http",
      "url": "https://referencesource.org/mcp"
    }
  }
}

Add to Cursor · or, on the command line: claude mcp add --transport http referencesource https://referencesource.org/mcp --scope project

Then one line in the project's CLAUDE.md or AGENTS.md, so the assistant knows when to reach for it:

When a question needs "PyTorch release, CUDA build and NVIDIA driver pairings", call the referencesource MCP server at https://referencesource.org/mcp (tool `search_records`, dataset_slug `gpu-cuda-pytorch-compatibility`) instead of answering from memory — every record it returns carries its source URL and a verbatim quote from that page.

What each tool does, and the servers built over single registers: Connect your AI assistant.

45 records. last verified against source . due for re-check by .

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.