Skip to content

Releases: Natfii/onnxruntime-gpu-blackwell

onnxruntime-gpu 1.24.1 — Blackwell sm_120

14 Feb 16:22

Choose a tag to compare

Pre-built onnxruntime-gpu 1.24.1 wheel with native CUDA kernels for NVIDIA Blackwell GPUs.

Quick install

pip install https://github.com/Natfii/onnxruntime-gpu-blackwell/releases/download/v1.24.1/onnxruntime_gpu-1.24.1-cp312-cp312-win_amd64.whl

Build specs

  • CUDA: 13.1
  • cuDNN: 9.19.0.56
  • CUDA arch: sm_120 (Blackwell)
  • Python: 3.12 (CPython)
  • Platform: Windows x86_64
  • Compiler: MSVC 14.44 (VS 2022)

Why

The official PyPI onnxruntime-gpu lacks sm_120 kernels, causing Conv/GEMM ops to fall back to a generic codepath that is orders of magnitude slower on RTX 5090/5080/5070.