Build and test with CUDA 13.0.0 #286
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
jameslamb
added
non-breaking
This comment was marked as resolved.
This comment was marked as resolved.
This was referenced Aug 26, 2025
jameslamb
commented
Sep 3, 2025
rapids-bot bot
pushed a commit
that referenced
this pull request
Sep 4, 2025
While trying to add CUDA 13 builds here, I discovered that this project's `build.sh` script has a hard-coded list of CUDA architectures: https://github.com/rapidsai/cugraph-gnn/blob/c50d56d0d4987c587e6f377dd4eb9537f3d0440b/build.sh#L212 As described in #286 (comment), this proposes removing that hard-coded list in favor of using the `"RAPIDS"` set of architectures from `rapids-cmake`. This also updates some `pre-commit` hooks versions (unrelated, but low-risk and wanted to take advantage of the CI runs). ## Notes for Reviewers ### Benefits of this change * keeps this project aligned with the rest of RAPIDS on CUDA architectures - reduces the manual effort required to support new CUDA versions - avoids the build time and binary size from building for older architectures that RAPIDS no longer supports ### So what architectures will `libwholegraph` now be built for? ```text # before 70-real;75-real;80-real;86-real;90 # after (CUDA 12) 70-real;75-real;80-real;86-real;90a-real;100f-real;120a-real;120 # after (CUDA 13) 75-real;80-real;86-real;90a-real;100f-real;120a-real;120 ``` Those lists come from here: https://github.com/rapidsai/rapids-cmake/blob/0b111489d1e6f8400e1fc88297623a2a9915fa77/rapids-cmake/cuda/set_architectures.cmake Authors: - James Lamb (https://github.com/jameslamb) Approvers: - Kyle Edwards (https://github.com/KyleFromNVIDIA) - https://github.com/linhu-nv URL: #295
jameslamb
changed the title
jameslamb
commented
Sep 5, 2025
This was referenced Sep 5, 2025
alexbarghi-nv
requested changes
Sep 5, 2025
Member
alexbarghi-nv
left a comment
alexbarghi-nv
left a comment
There was a problem hiding this comment.
What happens if we build with CUDA 13 but try to install the current PyTorch? Shouldn't PyT still work? Do they conflict on package versions? I'd rather we keep conda tests going if possible, plus we currently have a customer that's using nightly conda and I don't want them to get broken.
alexbarghi-nv
approved these changes
Sep 5, 2025
vyasr
approved these changes
Sep 5, 2025
Contributor
vyasr
left a comment
vyasr
left a comment
There was a problem hiding this comment.
Please open an issue with all the follow-up work associated with PyTorch and add it to our CUDA 13 rollout issue so that we know to keep track of that for our overall CUDA 13 rollout. Aside from that, LGTM, thanks!
Member
Author
|
/merge |
rapids-bot
bot
merged commit ef26ed9
into
rapidsai:branch-25.10
147 of 149 checks passed
Member
Author
|
Put up #296 documenting the need to add those tests, and linked it in the big task list on rapidsai/build-planning#208 |
rapids-bot bot
pushed a commit
to rapidsai/ci-imgs
that referenced
this pull request
Sep 5, 2025
Contributes to rapidsai/build-planning#208 Updates the `:latest` and `:25.10-latest` tags to CUDA 13.0.0. ## Notes for Reviewers ### is this safe to merge? Once these are in, I think yes: * [x] NVIDIA/cuopt#366 * [x] rapidsai/cugraph-gnn#286 At that point, the only thing it should affect are docs builds across repos that are already supporting CUDA 13 in all their other conda-based tests. Authors: - James Lamb (https://github.com/jameslamb) Approvers: - Kyle Edwards (https://github.com/KyleFromNVIDIA) URL: #303
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Contributes to rapidsai/build-planning#208
cuda-python:>=12.9.2(CUDA 12),>=13.0.1(CUDA 13)cupy:>=13.6.0Contributes to rapidsai/build-planning#68
dependencies.yamlmatrices (i.e., the ones that get written topyproject.tomlin source control)Notes for Reviewers
This switches GitHub Actions workflows to the
cuda13.0branch from here: rapidsai/shared-workflows#413A future round of PRs will revert that back to
branch-25.10, once all of RAPIDS supports CUDA 13.What about PyTorch?
There are now PyTorch CUDA 13 nightly wheels, but not yet conda packages.
CUDA 13 support is tracked in pytorch/pytorch#159779, and eventually will show up as PRs in https://github.com/conda-forge/pytorch-cpu-feedstock (ignore the feedstock name... that is really where the CUDA-enabled builds are too).
This PR proposes skipping CUDA 13 conda tests, so we can at least start producing nightly CUDA 13 packages here. If reviewers agree, I'll open an issue documenting the need to restore those tests.