cuda: fix check for GPU device availability #2510
Merged
Conversation
This file contains 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
rst0git
requested review from
jesus-ramos and
avagin
and removed request for
jesus-ramos
jesus-ramos
approved these changes
Nov 2, 2024
avagin
reviewed
Nov 4, 2024
rst0git
force-pushed
the
2024-11-02-cuda-check
branch
4 times, most recently
from
9a50892
to
a69ea00
Compare
avagin
reviewed
Nov 8, 2024
avagin
reviewed
Nov 8, 2024
|
LGTM. Thanks. |
rst0git
force-pushed
the
2024-11-02-cuda-check
branch
4 times, most recently
from
3d103d0
to
3940ee7
Compare
rst0git
changed the title
The check for `/dev/nvidiactl` to determine if the CUDA plugin can be used is unreliable because in some cases the default path for driver installation is different [1]. This patch changes the logic to check if a GPU device is available in `/proc/driver/nvidia/gpus/`. This approach is similar to `torch.cuda.is_available()` and it is a more accurate indicator. The subsequent check for support of the `cuda-checkpoint --action` option would confirm if the driver supports checkpoint/restore. [1] https://github.com/NVIDIA/gpu-operator Fixes: checkpoint-restore#2509 Signed-off-by: Radostin Stoyanov <rstoyanov@fedoraproject.org>
rst0git
force-pushed
the
2024-11-02-cuda-check
branch
from
3940ee7
to
de9d552
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
The check for
/dev/nvidiactlto determine if the CUDA plugin can be used is unreliable because in some cases the default path for driver installation is different 1. This pull request changes the logic to check if a GPU device is available in/proc/driver/nvidia/gpus/. This approach is similar totorch.cuda.is_available()and it is a more accurate indicator. The subsequent check for support of thecuda-checkpoint --actionoption would confirm if the driver supports checkpoint/restore.Fixes: #2509
Footnotes
https://github.com/NVIDIA/gpu-operator ↩