cuda.core: consolidate driver version query functions by leofang · Pull Request #2040 · NVIDIA/cuda-python

leofang · 2026-05-06T21:17:11Z

Summary

Consolidate get_driver_version() and get_driver_version_full() into a single get_driver_version() that returns (umd_version, kmd_version) — a pair of version tuples (UMD is a 2-tuple (MAJOR, MINOR), KMD is a 3-tuple (MAJOR, MINOR, PATCH)).

The function now requires NVML and raises RuntimeError if it is not available.

Removes the kernel_mode parameter and the separate get_driver_version_full function
Updates tests, docs, and 1.0.0 release notes

Merge the two separate driver version query functions into a single get_driver_version() that returns (umd_version, kmd_version) — a pair of version tuples. UMD is a 2-tuple (major, minor) and KMD is a 3-tuple (major, minor, patch). The function now requires NVML. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

copy-pr-bot · 2026-05-06T21:17:14Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

leofang · 2026-05-06T21:33:25Z

/ok to test bfd9ab9

github-actions · 2026-05-06T21:52:04Z

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-2040/
https://nvidia.github.io/cuda-python/pr-preview/pr-2040/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-2040/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-2040/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.

Windows driver version strings only have two components, so nvmlSystemGetDriverVersion returns X.Y on WSL instead of X.Y.Z. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

leofang · 2026-05-06T22:07:21Z

/ok to test d2d387e

rparolin · 2026-05-06T22:34:37Z

Nit: the file-level comment at _system.pyx#L6-L8 still states that functions must "fall back to non-NVML-based methods for backward compatibility" when cuda.bindings.nvml isn't available. After this PR, get_driver_version() no longer follows that pattern (it raises RuntimeError instead), joining get_driver_branch() / get_nvml_version() which already required NVML. Worth tweaking the header to reflect that the file is now a mix — some functions fall back (e.g. get_num_devices), others require NVML.

rparolin

The changes are correct but some added context would make it obvious to reader why its correct. I've labeled them as nit to not block this PR.

leofang · 2026-05-06T22:47:51Z

Let me wait for @mdboom blessing before merging

mdboom

I don't think we can make NVML a hard requirement, because it means we now require cuda_bindings >=12.9.6 or cuda_bindings >= 13.2.0. Don't we make broader guarantees in general for that?

To resolve that, maybe we just have two separate functions for kernel and cuda version, but remove the "_full" distinction (which was just there for backward compat).

…rsion Per review feedback, split the consolidated get_driver_version() into two separate functions so each returns a simple tuple[int, ...]: - get_user_mode_driver_version(): works with or without NVML (falls back to cuDriverGetVersion), returns (major, minor) - get_kernel_mode_driver_version(): requires NVML, returns (major, minor, patch) or (major, minor) on WSL Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

leofang · 2026-05-07T14:03:23Z

/ok to test f57003a

mdboom

LGTM

rwgk

Ah, looks like I was too slow.

rwgk · 2026-05-07T14:55:32Z

+    "get_kernel_mode_driver_version",
    "get_num_devices",
    "get_process_name",
+    "get_user_mode_driver_version",


For the cuda-pathfinder guard rails that I'm adding in pending PR #1977, I need both functions, too:

https://github.com/NVIDIA/cuda-python/pull/1977/changes#diff-e4eb8e0d6231c5ab21d8993a187523d4ef49b2b6eba39ee84b606968ea726722

Currently I have different naming. It might be good to keep the names aligned.

I was going with what nvidia-smi is showing, but that's changing. In the future it'll look like this:

+-----------------------------------------------------------------------------------------+ | NVIDIA-SMI NNN.NN KMD Version: NNN.NN CUDA UMD Version: NN.N | +-----------------------------------------+------------------------+----------------------+

So

get_driver_kmd_version
get_driver_cuda_umd_version

would be most aligned.

Removed preview folders for the following PRs: - PR #2040

github-actions Bot added the cuda.core Everything related to the cuda.core module label May 6, 2026

leofang self-assigned this May 6, 2026

leofang requested a review from mdboom May 6, 2026 21:18

leofang added enhancement Any code-related improvements P0 High priority - Must do! breaking Breaking changes are introduced labels May 6, 2026

leofang added this to the cuda.core v1.0.0 milestone May 6, 2026

Remove unused driver import from _system.pyx

bfd9ab9

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

leofang marked this pull request as ready for review May 6, 2026 21:33

Fix KMD version test to allow 2-tuple on WSL

d2d387e

Windows driver version strings only have two components, so nvmlSystemGetDriverVersion returns X.Y on WSL instead of X.Y.Z. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

rparolin reviewed May 6, 2026

View reviewed changes

Comment thread cuda_core/tests/system/test_system_system.py Outdated

rparolin reviewed May 6, 2026

View reviewed changes

Comment thread cuda_core/tests/system/test_system_system.py

rparolin approved these changes May 6, 2026

View reviewed changes

mdboom requested changes May 7, 2026

View reviewed changes

leofang and others added 5 commits May 7, 2026 13:45

Add example version strings to driver version docstrings

4c7cd2c

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Remove NVML implementation detail from UMD docstring

27ad775

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Sort __all__ alphabetically in system/__init__.py

2fe17c8

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Merge branch 'main' into consolidate-driver-version

f57003a

leofang requested a review from mdboom May 7, 2026 14:03

mdboom approved these changes May 7, 2026

View reviewed changes

leofang enabled auto-merge (squash) May 7, 2026 14:31

leofang merged commit 57a538c into NVIDIA:main May 7, 2026
94 checks passed

rwgk reviewed May 7, 2026

View reviewed changes

leofang deleted the consolidate-driver-version branch May 7, 2026 15:06

github-actions Bot pushed a commit that referenced this pull request May 8, 2026

Clean up PR preview folders for 1 closed/merged PRs

412f4f6

Removed preview folders for the following PRs: - PR #2040

Conversation

leofang commented May 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Uh oh!

copy-pr-bot Bot commented May 6, 2026

Uh oh!

leofang commented May 6, 2026

Uh oh!

github-actions Bot commented May 6, 2026

Preview will be ready when the GitHub Pages deployment is complete.

Uh oh!

leofang commented May 6, 2026

Uh oh!

Uh oh!

Uh oh!

rparolin commented May 6, 2026

Uh oh!

rparolin left a comment

Choose a reason for hiding this comment

Uh oh!

leofang commented May 6, 2026

Uh oh!

mdboom 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!

leofang commented May 7, 2026

Uh oh!

mdboom left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

rwgk left a comment

Choose a reason for hiding this comment

Uh oh!

rwgk May 7, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

leofang commented May 6, 2026 •

edited

Loading

mdboom left a comment •

edited

Loading