Fix CUDA V100 compatibility with LocalPreferences.toml#868
Merged
ChrisRackauckas merged 7 commits intoSciML:masterfrom Mar 20, 2026
Merged
Conversation
This fixes the CUDA test failures on V100 runners (demeter3/demeter4) by: 1. Adding LocalPreferences.toml files to pin CUDA Runtime to v12.6 and disable the forward-compat driver (CUDA_Driver_jll v13+ drops V100 compute capability 7.0 support) 2. Updating test runtests.jl to add CUDA_Driver_jll and CUDA_Runtime_jll packages so the preferences are properly loaded Fixes: ChrisRackauckas/InternalJunk#24
…ation ForwardDiff uses scalar indexing which triggers GPU scalar indexing errors. Use AutoFiniteDiff which works with CUDA arrays.
The default PolyAlgorithm uses AutoForwardDiff internally which does scalar indexing on GPU arrays. Since there's no way to pass autodiff to the default algorithm, remove it from GPU tests. Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com> Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com> Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add `needs: nonlinearsolve-cuda` to SimpleNonlinearSolve CUDA job so they don't compete for GPU memory on the same V100 runner - Add GC.gc() and CUDA.reclaim() before kernel launch tests to free memory from prior test items on shared GPU runners Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com> Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The `needs` dependency doesn't help since other jobs on the shared runner contribute to GPU memory pressure regardless. The CUDA.reclaim() fix in the test file is the actual mitigation. Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com> Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The OOM was transient (other processes on shared runner), not caused by our tests. The gpu-v100 tag already targets the right runners. Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com> Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
Summary
This PR fixes the CUDA test failures on V100 runners (demeter3/demeter4) that were caused by CUDA driver version compatibility issues.
Problem
The V100 GPUs on demeter4 have a CUDA driver version (570.172) that reports an NVML version mismatch with newer CUDA toolkit versions. Additionally,
CUDA_Driver_jllv13+ drops support for compute capability 7.0 (V100).The tests were failing with:
Solution
Following the pattern established in OrdinaryDiffEq.jl PR #3162:
Added
LocalPreferences.tomlfiles in both the root directory andlib/SimpleNonlinearSolve/to:compat = "false")Updated
test/runtests.jlandlib/SimpleNonlinearSolve/test/runtests.jlto addCUDA_Driver_jllandCUDA_Runtime_jllpackages when running CUDA tests, so the LocalPreferences are properly loaded.Files Changed
LocalPreferences.toml(new)lib/SimpleNonlinearSolve/LocalPreferences.toml(new)test/runtests.jllib/SimpleNonlinearSolve/test/runtests.jlReferences