{
  "dataset": "tensorflow",
  "source": "https://gpucompat.com/data/tensorflow.json",
  "about": "https://gpucompat.com/about-the-data/",
  "count": 17,
  "entries": [
    {
      "id": "2-10-0",
      "version": "2.10.0",
      "releaseDate": "2022-09-06",
      "cudaVersion": "11.2",
      "cudnnVersion": "8.1",
      "pythonVersions": [
        "3.7",
        "3.8",
        "3.9",
        "3.10"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options. Last TensorFlow release with native Windows GPU support -- TensorFlow 2.11.0 onward dropped native-Windows CUDA builds; Windows users need WSL2 (a Linux CUDA build) instead."
    },
    {
      "id": "2-11-0",
      "version": "2.11.0",
      "releaseDate": "2022-11-18",
      "cudaVersion": "11.2",
      "cudnnVersion": "8.1",
      "pythonVersions": [
        "3.7",
        "3.8",
        "3.9",
        "3.10"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options. First release without native Windows GPU support (see the 2.10.0 entry) -- Windows users must use WSL2."
    },
    {
      "id": "2-12-0",
      "version": "2.12.0",
      "releaseDate": "2023-03-22",
      "cudaVersion": "11.8",
      "cudnnVersion": "8.6",
      "pythonVersions": [
        "3.8",
        "3.9",
        "3.10",
        "3.11"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-13-0",
      "version": "2.13.0",
      "releaseDate": "2023-07-05",
      "cudaVersion": "11.8",
      "cudnnVersion": "8.6",
      "pythonVersions": [
        "3.8",
        "3.9",
        "3.10",
        "3.11"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-14-0",
      "version": "2.14.0",
      "releaseDate": "2023-09-26",
      "cudaVersion": "11.8",
      "cudnnVersion": "8.7",
      "pythonVersions": [
        "3.9",
        "3.10",
        "3.11"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-15-0",
      "version": "2.15.0",
      "releaseDate": "2023-11-14",
      "cudaVersion": "12.2",
      "cudnnVersion": "8.9",
      "pythonVersions": [
        "3.9",
        "3.10",
        "3.11"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-16-1",
      "version": "2.16.1",
      "releaseDate": "2024-03-09",
      "cudaVersion": "12.3",
      "cudnnVersion": "8.9",
      "pythonVersions": [
        "3.9",
        "3.10",
        "3.11",
        "3.12"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options. 2.16.1 is the first 2.16.x release with a published GPU build config in TensorFlow's own tested-configurations table (2.16.0 has no corresponding row there)."
    },
    {
      "id": "2-17-0",
      "version": "2.17.0",
      "releaseDate": "2024-07-11",
      "cudaVersion": "12.3",
      "cudnnVersion": "8.9",
      "pythonVersions": [
        "3.9",
        "3.10",
        "3.11",
        "3.12"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-18-0",
      "version": "2.18.0",
      "releaseDate": "2024-10-25",
      "cudaVersion": "12.5",
      "cudnnVersion": "9.3",
      "pythonVersions": [
        "3.9",
        "3.10",
        "3.11",
        "3.12"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-19-0",
      "version": "2.19.0",
      "releaseDate": "2025-03-12",
      "cudaVersion": "12.5",
      "cudnnVersion": "9.3",
      "pythonVersions": [
        "3.9",
        "3.10",
        "3.11",
        "3.12"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-20-0",
      "version": "2.20.0",
      "releaseDate": "2025-08-13",
      "cudaVersion": "12.5",
      "cudnnVersion": "9.3",
      "pythonVersions": [
        "3.9",
        "3.10",
        "3.11",
        "3.12",
        "3.13"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-21-0",
      "version": "2.21.0",
      "releaseDate": "2026-03-06",
      "cudaVersion": "12.5",
      "cudnnVersion": "9.3",
      "pythonVersions": [
        "3.10",
        "3.11",
        "3.12",
        "3.13"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-5-0",
      "version": "2.5.0",
      "releaseDate": "2021-05-13",
      "cudaVersion": "11.2",
      "cudnnVersion": "8.1",
      "pythonVersions": [
        "3.6",
        "3.7",
        "3.8",
        "3.9"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "Added 2026-09 at the user's request to extend coverage back past the existing 2.7.0 floor. CUDA/cuDNN pin and Python range verified against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Same CUDA/cuDNN pin as 2.6.0 and 2.7.0 -- TensorFlow held 11.2/8.1 across all three releases."
    },
    {
      "id": "2-6-0",
      "version": "2.6.0",
      "releaseDate": "2021-08-11",
      "cudaVersion": "11.2",
      "cudnnVersion": "8.1",
      "pythonVersions": [
        "3.6",
        "3.7",
        "3.8",
        "3.9"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "Added 2026-09 alongside 2.5.0 at the user's request. CUDA/cuDNN pin and Python range verified against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Same CUDA/cuDNN pin as 2.5.0 and 2.7.0 -- TensorFlow held 11.2/8.1 across all three releases."
    },
    {
      "id": "2-7-0",
      "version": "2.7.0",
      "releaseDate": "2021-11-04",
      "cudaVersion": "11.2",
      "cudnnVersion": "8.1",
      "pythonVersions": [
        "3.7",
        "3.8",
        "3.9"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-8-0",
      "version": "2.8.0",
      "releaseDate": "2022-02-02",
      "cudaVersion": "11.2",
      "cudnnVersion": "8.1",
      "pythonVersions": [
        "3.7",
        "3.8",
        "3.9",
        "3.10"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    },
    {
      "id": "2-9-0",
      "version": "2.9.0",
      "releaseDate": "2022-05-16",
      "cudaVersion": "11.2",
      "cudnnVersion": "8.1",
      "pythonVersions": [
        "3.7",
        "3.8",
        "3.9",
        "3.10"
      ],
      "rocmNote": "TensorFlow's official pip wheels are CUDA-only; there is no NVIDIA/ROCm choice like PyTorch's. ROCm support exists via AMD's separate tensorflow-rocm build, which follows its own versioning and isn't tracked here.",
      "notes": "CUDA/cuDNN pin and Python range verified 2026-08 against TensorFlow's own \"tested build configurations\" table at tensorflow.org/install/source#gpu (Linux GPU config); release date from PyPI upload metadata. Unlike PyTorch, TensorFlow's official pip wheels are built against exactly one CUDA/cuDNN pair per release, not a matrix of options."
    }
  ]
}