Ensure CUDA devices are usable before training#585
Open
Conversation
rob-maron
changed the title
ethernet8023
approved these changes
Feb 25, 2026
Contributor
ethernet8023
left a comment
ethernet8023
left a comment
There was a problem hiding this comment.
This LGTM. I'm not worried about RTX20xx and older devices for now, I imagine we have a hundred other compatibility issues with them. We can return to this later if we decide to start supporting that hardware.
Collaborator
Author
|
Did some testing on some different cards. As expected, it fails on T4s (Turing), but succeeds on H100s, A100s, L4s |
rob-maron
force-pushed
the
cuda-smoke-test
branch
2 times, most recently
from
8d424fc to
0363f80
Compare
rob-maron
force-pushed
the
cuda-smoke-test
branch
from
cd16397 to
1cea15b
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Closes #550
Adds a CUDA smoke test that validates that the requested devices are functional before training begins
The test is designed to trigger cuDNN runtime compilation to catch issues early. Credit to @arilotter for the core test logic.
One question potentially for discussion is whether or not this gates out older cards. I think it should run fine on even smaller Turing cards, but anything older it might not