Set UCX_CUDA_COPY_MAX_REG_RATIO=1.0 when GPUs permit #824
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
This is necessary to maintain good performance on GPUs with large BAR1 size (greater or equal to the total GPU memory size). Starting with UCX 1.12 this value is set to 0.1 by default to ensure registration doesn't fail with GPUs that have a BAR1 size lower than the total GPU memory size (e.g., T4). Currently we're setting to 1.0 when all GPUs have a large BAR1, in the future it may be useful to compute the real BAR1/TotalGPUMemory ratio to set this number. This is also a global setting, for fine-grained settings the application should take care of changing this value appropriately as UCX-Py doesn't know which GPUs are going to be used by the application at import time.
|
Thank you tracking this down @pentschev ! |
quasiben
approved these changes
Jan 3, 2022
pentschev
changed the title
|
rerun tests |
|
Tested and confirmed this fixed a performance issue -- merging in. Thank you again @pentschev ! |
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Outdated suggestions cannot be applied.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.
This is necessary to maintain good performance on GPUs with large BAR1 size (greater or equal to the total GPU memory size). Starting with UCX 1.12 this value is set to 0.1 by default to ensure registration doesn't fail with GPUs that have a BAR1 size lower than the total GPU memory size (e.g., T4).
Currently we're setting to 1.0 when all GPUs have a large BAR1, in the future it may be useful to compute the real BAR1/TotalGPUMemory ratio to set this number. This is also a global setting, for fine-grained settings the application should take care of changing this value appropriately as UCX-Py doesn't know which GPUs are going to be used by the application at import time.
Additionally, setup logger before environment variables are set and log what's being set.