File size: 7,867 Bytes
7644d36 b91f8bd f75f7c8 7644d36 f75f7c8 7644d36 f75f7c8 7644d36 f75f7c8 7644d36 f75f7c8 7644d36 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 | ---
pretty_name: "PyTorch release, CUDA build and NVIDIA driver pairings"
license: other
license_name: source-terms-cited
license_link: https://referencesource.org/gpu-cuda-pytorch-compatibility/
language:
- en
tags:
- reference-data
- provenance
- referencesource-org
configs:
- config_name: default
data_files:
- split: train
path: data/records.jsonl
---
# PyTorch release, CUDA build and NVIDIA driver pairings
**Canonical, always-current version: https://referencesource.org/gpu-cuda-pytorch-compatibility/**
Machine-readable: https://referencesource.org/gpu-cuda-pytorch-compatibility/data.json — this mirror is a point-in-time copy.
- **Last verified:** 2026-09-29
- **Stale after:** 2026-12-28 (past this date, prefer the canonical copy —
it re-verifies on a cadence this snapshot does not)
- **Records:** 45
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.
## Provenance — what every record carries
Every record carries `source` (the URL the value was read from) and
`source_quote` (a verbatim quote from that page stating it). A value you
cannot check against its page is indistinguishable from an invented one;
these can all be checked. Each record is also individually addressable on
the canonical site at its `url`.
## Licence position
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.
Facts are not copyrightable; what this dataset takes from each source is a
fact plus a short attributed quote linking back to the page that states it.
## 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/2.11/RELEASE.md
- 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/RELEASE.md
- https://raw.githubusercontent.com/pytorch/pytorch/release/2.12/.ci/manywheel/build_cuda.sh
- 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/RELEASE.md
- https://raw.githubusercontent.com/pytorch/pytorch/release/2.13/.ci/pytorch/windows/internal/cuda_config.bat
- https://raw.githubusercontent.com/pytorch/pytorch/release/2.14/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://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
## Fields
See any record in `data/records.jsonl` — field names are self-describing,
and the canonical page for each record renders them with labels:
https://referencesource.org/gpu-cuda-pytorch-compatibility/
|