[air] Increase global batch size for air_benchmark_tensorflow_mnist_gpu_4x4 #31402
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
…pu_4x4 Signed-off-by: Kai Fricke <kai@anyscale.com>
krfricke
changed the title
xwjiang2010
approved these changes
Jan 3, 2023
|
Increasing the batch size does not increase GPU util - likely because the dataset size is limited (fashion nist only has 60k examples). For that we would have to add more data |
AmeerHajAli
pushed a commit
that referenced
this pull request
Jan 12, 2023
…pu_4x4 (#31402) The benchmark currently times out, because a single run takes over 16 minutes. This is a regression compared to e.g. the 2.0.0 release, where a run took only 4 minutes. Upon closer investigation, this seems to be related to severe underutilization of the GPU. With a small batch size, we are bound by compute/data iteration, and this seems to have been regressed in later tensorflow versions. To achieve shorter training times, we increase the global batch size from 64 (which is tiny) to 1024. This severely speeds up training (even though the GPUs are still underutilized with <10% utilizaton). Signed-off-by: Kai Fricke <kai@anyscale.com>
tamohannes
pushed a commit
to ju2ez/ray
that referenced
this pull request
Jan 25, 2023
…pu_4x4 (ray-project#31402) The benchmark currently times out, because a single run takes over 16 minutes. This is a regression compared to e.g. the 2.0.0 release, where a run took only 4 minutes. Upon closer investigation, this seems to be related to severe underutilization of the GPU. With a small batch size, we are bound by compute/data iteration, and this seems to have been regressed in later tensorflow versions. To achieve shorter training times, we increase the global batch size from 64 (which is tiny) to 1024. This severely speeds up training (even though the GPUs are still underutilized with <10% utilizaton). Signed-off-by: Kai Fricke <kai@anyscale.com> Signed-off-by: tmynn <hovhannes.tamoyan@gmail.com>
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.
Signed-off-by: Kai Fricke kai@anyscale.com
Why are these changes needed?
The benchmark currently times out, because a single run takes over 16 minutes. This is a regression compared to e.g. the 2.0.0 release, where a run took only 4 minutes.
Upon closer investigation, this seems to be related to severe underutilization of the GPU. With a small batch size, we are bound by compute/data iteration, and this seems to have been regressed in later tensorflow versions.
To achieve shorter training times, we increase the global batch size from 64 (which is tiny) to 1024. This severely speeds up training (even though the GPUs are still underutilized with <10% utilizaton).
Related issue number
Closes #29922
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.