[GPU] Copy to host in case of small matrices to release resources early #136464
+88
−36
Conversation
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
ldematte
added
>non-issue
:Search Relevance/Vectors
|
Pinging @elastic/es-search-relevance (Team:Search Relevance) |
elasticsearchmachine
added
the
Team:Search Relevance
|
Update without test-gpu to verify this works well with "regular" CI |
georgewallace
pushed a commit
to georgewallace/elasticsearch
that referenced
this pull request
Oct 13, 2025
…ly (elastic#136464) This PR makes a small change to improve parallelism during graph build which we noticed with NVIDIA from profiler traces. In case the resulting graph is "small enough" (where "small enough" is ATM set to 128 MB) we copy the graph entirely to host memory, release the cuvs resources and proceed, instead of downloading data in pages from the device and write each page to disk, which is more efficient but will hold the resources till we finished writing to disk -- on a busy system this can require time.
ldematte
added a commit
to ldematte/elasticsearch
that referenced
this pull request
Oct 14, 2025
…ly (elastic#136464) This PR makes a small change to improve parallelism during graph build which we noticed with NVIDIA from profiler traces. In case the resulting graph is "small enough" (where "small enough" is ATM set to 128 MB) we copy the graph entirely to host memory, release the cuvs resources and proceed, instead of downloading data in pages from the device and write each page to disk, which is more efficient but will hold the resources till we finished writing to disk -- on a busy system this can require time.
elasticsearchmachine
pushed a commit
that referenced
this pull request
Oct 14, 2025
…ly (#136464) (#136532) This PR makes a small change to improve parallelism during graph build which we noticed with NVIDIA from profiler traces. In case the resulting graph is "small enough" (where "small enough" is ATM set to 128 MB) we copy the graph entirely to host memory, release the cuvs resources and proceed, instead of downloading data in pages from the device and write each page to disk, which is more efficient but will hold the resources till we finished writing to disk -- on a busy system this can require time.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
This PR makes a small change to improve parallelism during graph build which we noticed with NVIDIA from profiler traces.
In case the resulting graph is "small enough" (where "small enough" is ATM set to 128 MB) we copy the graph entirely to host memory, release the cuvs resources and proceed, instead of downloading data in pages from the device and write each page to disk, which is more efficient but will hold the resources till we finished writing to disk -- on a busy system this can require time.