Skip to content

[CUDA] Add nightly gpu package and publish pipelines for CUDA 13 - #26420

Merged
Tianlei Wu (tianleiwu) merged 3 commits into
mainfrom
tlwu/cuda130_nightly_package_pipelines
Oct 28, 2025
Merged

[CUDA] Add nightly gpu package and publish pipelines for CUDA 13#26420
Tianlei Wu (tianleiwu) merged 3 commits into
mainfrom
tlwu/cuda130_nightly_package_pipelines

Conversation

@tianleiwu

Copy link
Copy Markdown
Contributor

Add nightly gpu pipelines for CUDA 13.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull Request Overview

This PR adds support for CUDA 13 to the nightly GPU package build and publishing pipelines. The changes enable building, testing, and publishing Python packages with CUDA 13.0 support alongside the existing CUDA 12.8 infrastructure.

Key Changes:

  • Added CUDA 13.0 as a valid option across multiple pipeline configuration files
  • Created dedicated CUDA 13 packaging and publishing pipeline definitions
  • Refactored Windows GPU library download logic to support CUDA 13-specific dependencies (CUDNN and TensorRT)

Reviewed Changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file
File Description
py-packaging-linux-test-cuda.yml Added CUDA 13.0 to allowed CudaVersion values
download_win_gpu_library.yml Refactored TensorRT parameter handling and added CUDNN download support for CUDA 13
common-variables.yml Fixed Windows TensorRT folder naming convention for CUDA 13
py-win-gpu-stage.yml Added CUDA 13.0 to allowed CudaVersion values
py-linux-gpu-stage.yml Added CUDA 13.0 to allowed CudaVersion values
py-gpu-packaging-stage.yml Added conditional logic for CUDA 13 builds with separate configuration
py-cuda13-publishing-pipeline.yml Created new pipeline for publishing CUDA 13 packages
py-cuda13-packaging-pipeline.yml Created new pipeline for building CUDA 13 packages
onnxruntime_providers_cuda.cmake Added /bigobj compiler flag for MSVC CUDA compilation

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

@tianleiwu
Tianlei Wu (tianleiwu) merged commit 1633299 into main Oct 28, 2025
110 of 122 checks passed
@tianleiwu
Tianlei Wu (tianleiwu) deleted the tlwu/cuda130_nightly_package_pipelines branch October 28, 2025 16:28
@btakita

Brian Takita (btakita) commented Oct 28, 2025

Copy link
Copy Markdown

Thank you for your work to get this done & merged! Should there be instructions to installing the nightly Cuda 13.0 version at https://onnxruntime.ai/docs/install/?

Naomi Ovadia (naomiOvad) pushed a commit to naomiOvad/onnxruntime that referenced this pull request Nov 2, 2025
Tianlei Wu (tianleiwu) added a commit that referenced this pull request Nov 4, 2025
Update python packaging pipeline to support python 3.14. This is to add
python 3.14 wheels to both CUDA 12 and CUDA 13 nightly packaging
pipelines. The CUDA 13 pipelines were added in
#26420.

For linux, two wheels are added for python 3.14 and 3.14t.
For Windows, a wheel is added for pyhthon 3.14.

Known issues:
1. onnx does not have prebuild package for python 3.14, which causes
some pipelines failed.
2.  onnxscript and onnx-ir depends on onnx.
3. The CI machine for Windows cuda python packaging has cuda driver
553.24, which does not support CUDA 13.0. TODO: Update cuda driver to
580 or later to test properly.

Walkarounds for issues 1 and 2:
* For python 3.14, we skip onnx, onnxscript and onnx-ir in
requirements.txt.
* Disable related tests.

When onnx package supports python 3.14 in the future, we can undo the
walkarounds and re-enable those tests.
Rohanjames1997 pushed a commit to Rohanjames1997/onnxruntime that referenced this pull request Dec 4, 2025
Update python packaging pipeline to support python 3.14. This is to add
python 3.14 wheels to both CUDA 12 and CUDA 13 nightly packaging
pipelines. The CUDA 13 pipelines were added in
microsoft#26420.

For linux, two wheels are added for python 3.14 and 3.14t.
For Windows, a wheel is added for pyhthon 3.14.

Known issues:
1. onnx does not have prebuild package for python 3.14, which causes
some pipelines failed.
2.  onnxscript and onnx-ir depends on onnx.
3. The CI machine for Windows cuda python packaging has cuda driver
553.24, which does not support CUDA 13.0. TODO: Update cuda driver to
580 or later to test properly.

Walkarounds for issues 1 and 2:
* For python 3.14, we skip onnx, onnxscript and onnx-ir in
requirements.txt.
* Disable related tests.

When onnx package supports python 3.14 in the future, we can undo the
walkarounds and re-enable those tests.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants