Build and test with CUDA 13.0.0 by jameslamb · Pull Request #803 · rapidsai/kvikio

jameslamb · 2025-08-19T15:21:30Z

Contributes to rapidsai/build-planning#208

uses CUDA 13.0.0 to build and test

Contributes to rapidsai/build-planning#68

updates to CUDA 13 dependencies in fallback entries in dependencies.yaml matrices (i.e., the ones that get written to pyproject.toml in source control)

Notes for Reviewers

This switches GitHub Actions workflows to the cuda13.0 branch from here: rapidsai/shared-workflows#413

A future round of PRs will revert that back to branch-25.10, once all of RAPIDS supports CUDA 13.

copy-pr-bot · 2025-08-19T15:21:33Z

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

jameslamb · 2025-08-19T16:08:28Z

/ok to test

jameslamb · 2025-08-19T16:53:22Z

Problem 1: nvCOMP 4.2.0.11 does not have CUDA 13 packages

Python conda builds are failing like this:

Error:   × Failed to resolve dependencies: Cannot solve the request because of:
  │ libnvcomp-dev 4.2.0.11.* cannot be installed because there are no viable
  │ options:
  │ └─ libnvcomp-dev 4.2.0.11 would require
  │    └─ cuda-version >=12,<13.0a0, for which no candidates were found.
  │ 
  ╰─▶ Cannot solve the request because of: libnvcomp-dev 4.2.0.11.* cannot be
      installed because there are no viable options:
      └─ libnvcomp-dev 4.2.0.11 would require
         └─ cuda-version >=12,<13.0a0, for which no candidates were found.

(build link)

Luckily #804 will remove the dependency on nvCOMP here, which should resolve that.

Problem 2: failing Python tests

There are a bunch of failing Python tests, concentrated on many different parameterizations of 2 test cases.

FAILED python/kvikio/tests/test_mmap.py::test_read_seq[cupy-0-None] - TypeError: bad any_cast
FAILED python/kvikio/tests/test_mmap.py::test_read_seq[cupy-0-10] - TypeError: bad any_cast
FAILED python/kvikio/tests/test_mmap.py::test_read_seq[cupy-0-9999] - TypeError: bad any_cast
...
FAILED python/kvikio/tests/test_mmap.py::test_read_with_default_arguments[cupy] - TypeError: bad any_cast
FAILED python/kvikio/tests/test_mmap.py::test_read_with_default_arguments[cupy_async] - TypeError: bad any_cast
================= 92 failed, 1404 passed, 13 skipped in 59.70s =================

(build link)

jameslamb · 2025-08-19T17:58:28Z

/ok to test

jameslamb · 2025-08-19T18:10:10Z

/ok to test

kingcrimsontianyu · 2025-08-20T05:21:35Z

/ok to test aa320b5

bdice

Requesting a few related version bumps that should ensure a smoother CUDA 13 experience.

cpp/doxygen/main_page.md

bdice · 2025-08-20T12:20:31Z

cpp/src/shim/cuda.cpp

-    decltype(cuMemcpyBatchAsync)* fp;
-    get_symbol(fp, lib, KVIKIO_STRINGIFY(cuMemcpyBatchAsync));
-    MemcpyBatchAsync.set(fp);
+    if ((CUDA_VERSION >= 12080 && CUDA_VERSION < 13000 && driver_version >= 12080 &&


I feel like there must be a way for us to support CUDA 12.8 runtime with CUDA 13.0 driver. Can we investigate this a little further?

Many of our CUDA 12 tests are running with driver 580, which should cover this case.

Thanks.
It looks like because of backward compatibility the code compiled against CUDA driver 12.9 is able to dynamically load its own version of batched copy function from libcuda.so.580.65.06 when run on the system with CUDA driver 13.0.
Handling simplified.

bdice · 2025-08-20T12:22:35Z

dependencies.yaml

      - output_types: conda
        packages:
-          - cupy>=12.0.0
+          - &cupy_unsuffixed cupy>=12.0.0


We need to bump to cupy>=13.6 for CUDA 13. Maybe we should require that same pinning on CUDA 12, too? Historically I believe we've bumped cupy bounds for all CUDA versions whenever some new CUDA version needs a higher pinning.

I don't think we are testing cupy==12.0.0 in our "oldest" dependencies jobs right now so I don't have a good idea of whether 12.0 still works. Bumping to 13.6 seems safest.

Ok thanks for that historical context.

On https://anaconda.org/conda-forge/cupy/files?sort=basename&sort_order=desc&version=13.6.0 I do see cupy==13.6.0 packages published for CUDA 11, 12, and 13, so hopefully this should be ok.

Updated in 6e7aff3, and I'll update the other CUDA 13 PRs (including rapids-reviser example).

dependencies.yaml

docs/source/install.rst

jameslamb · 2025-08-20T14:14:13Z

At this point everyone involved in the PR is already subscribed to notifications via comments, so I've moved this out of draft so that we don't have to /ok to test new commits.

bdice

One small docs suggestion, otherwise LGTM

bdice · 2025-08-20T19:37:08Z

.devcontainer/cuda13.0-conda/devcontainer.json

@jameslamb Just as a heads-up, make sure the newly created devcontainers include Paul's build cluster changes. I think this looks right but just verify that for all the repos as you add CUDA 13 support to them.

Good note, thank you. I think we have what's needed here, based on what I see in the diff from #797

The only changes in the devcontainer.json files themselves were things like this, which this branch has:

"runArgs": [ "--rm", "--name", "${localEnv:USER:anon}-rapids-${localWorkspaceFolderBasename}-25.10-cuda13.0-conda", + "--ulimit", + "nofile=500000" ],

https://github.com/jameslamb/kvikio/blob/3162121ad72d303b150e7e80a0e61c0bac06fd3b/.devcontainer/cuda13.0-conda/devcontainer.json#L15-L16

docs/source/install.rst

Co-authored-by: Bradley Dice <bdice@bradleydice.com>

jameslamb · 2025-08-20T21:02:45Z

/merge

Build and test with CUDA 13.0.0

c96616e

jameslamb added non-breaking Introduces a non-breaking change improvement Improves an existing functionality labels Aug 19, 2025

fix dependencies, update docs

b3ca457

Merge branch 'branch-25.10' into cuda-13.0.0

e0bfabe

dependencies.yaml

12286bc

jameslamb mentioned this pull request Aug 19, 2025

Add CUDA 13.0 Support across all of RAPIDS rapidsai/build-planning#208

Closed

Attempt to fix the failed unit test in CUDA 13 due to batched memcpy

aa320b5

bdice reviewed Aug 20, 2025

View reviewed changes

jameslamb added 3 commits August 20, 2025 09:01

Merge branch 'branch-25.10' into cuda-13.0.0

3f7349a

add devcontainers

c909012

update cuda-python and cupy pins, re-arrange docs

6e7aff3

jameslamb changed the title ~~WIP: Build and test with CUDA 13.0.0~~ Build and test with CUDA 13.0.0 Aug 20, 2025

jameslamb requested a review from kingcrimsontianyu August 20, 2025 14:13

jameslamb marked this pull request as ready for review August 20, 2025 14:13

jameslamb requested review from a team as code owners August 20, 2025 14:13

jameslamb requested a review from bdice August 20, 2025 14:13

kingcrimsontianyu added 2 commits August 20, 2025 11:09

Update and simplify CUDA version handling

ca48989

Merge remote-tracking branch 'james-origin/cuda-13.0.0' into cuda-13.0.0

4cafa8d

bdice approved these changes Aug 20, 2025

View reviewed changes

Update docs/source/install.rst

3162121

Co-authored-by: Bradley Dice <bdice@bradleydice.com>

rapids-bot bot merged commit c774079 into rapidsai:branch-25.10 Aug 20, 2025
77 checks passed

jameslamb deleted the cuda-13.0.0 branch August 20, 2025 21:02

jameslamb mentioned this pull request Aug 21, 2025

Build and test with CUDA 13.0.0 rapidsai/ucx-py#1162

Merged

jameslamb mentioned this pull request Sep 9, 2025

Standardize dependency pins for RAPIDS 25.10 rapidsai/build-planning#214

Closed

17 tasks

Conversation

jameslamb commented Aug 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Notes for Reviewers

Uh oh!

copy-pr-bot bot commented Aug 19, 2025

Uh oh!

jameslamb commented Aug 19, 2025

Uh oh!

jameslamb commented Aug 19, 2025

Problem 1: nvCOMP 4.2.0.11 does not have CUDA 13 packages

Problem 2: failing Python tests

Uh oh!

jameslamb commented Aug 19, 2025

Uh oh!

jameslamb commented Aug 19, 2025

Uh oh!

kingcrimsontianyu commented Aug 20, 2025

Uh oh!

bdice left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

bdice Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

bdice Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

kingcrimsontianyu Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

bdice Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

jameslamb Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jameslamb commented Aug 20, 2025

Uh oh!

bdice left a comment

Choose a reason for hiding this comment

Uh oh!

bdice Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

jameslamb Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jameslamb commented Aug 20, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

jameslamb commented Aug 19, 2025 •

edited

Loading

bdice left a comment •

edited

Loading