Added requested_bytes to CUDA Caching Allocator Stats #88575
Closed
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 pull request was exported from Phabricator. Differential Revision: D40810674 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Nov 7, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: 7c6d71e6c307d87d6b6b5db5d45d1e9b5b354f42
c-odrin
force-pushed
the
export-D40810674
branch
from
f6b245f
to
3bd959a
Compare
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Nov 7, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: 6cd3715887a58b567b759ecd063d46cca3376777
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
c-odrin
force-pushed
the
export-D40810674
branch
from
3bd959a
to
609851b
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
c-odrin
force-pushed
the
export-D40810674
branch
from
609851b
to
7f16c56
Compare
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Nov 7, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: 9aa71acd15755d6fbec17a61ca6fcfb69a7d2380
c-odrin
force-pushed
the
export-D40810674
branch
from
7f16c56
to
050cba1
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Nov 9, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: 5f2719eef86fe6e5158f1f330cad6b69a981d183
c-odrin
force-pushed
the
export-D40810674
branch
from
050cba1
to
9b98296
Compare
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Nov 10, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: 6b4da60ead445ee095904a0c9cfade771b66409d
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
c-odrin
force-pushed
the
export-D40810674
branch
from
9b98296
to
e80ba76
Compare
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Nov 11, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: 3b962b0b44f482a5d0d71c177e6bab83d9b9d887
zdevito
reviewed
Dec 5, 2022
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
c-odrin
force-pushed
the
export-D40810674
branch
from
e80ba76
to
63f645d
Compare
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Dec 6, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: 297b81c7886f5725a131848288652b35cdbcbd0d
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Dec 6, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: e134240786e9a0cbf06d806f61adde57bf78a0c5
c-odrin
force-pushed
the
export-D40810674
branch
from
63f645d
to
348f39f
Compare
c-odrin
force-pushed
the
export-D40810674
branch
from
3f07abe
to
4e65787
Compare
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Dec 11, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: 44a9fcd7bbf8971f0e8acc52f23e24ddc9ae5e42
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Dec 11, 2022
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: fbc132e4b66f49a650d5dde2fd040752109d5c85
c-odrin
force-pushed
the
export-D40810674
branch
from
4e65787
to
1f0e522
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
Thank you for the review! I've addressed the inline comments, added tests for reading requested_bytes out of segment info/block info and added code to reset the statistics when all the other statistics are reset. |
zdevito
approved these changes
Dec 13, 2022
c-odrin
force-pushed
the
export-D40810674
branch
from
1f0e522
to
22f94d8
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
c-odrin
added a commit
to c-odrin/pytorch
that referenced
this pull request
Jan 10, 2023
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: 06dbdd60a76d80a1bfdeb6cbbacba9af1e7a1f53
Summary: Pull Request resolved: pytorch#88575 The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce. We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag: - "requested_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead Test Plan: Added test case in caffe2/test/test_cuda.py Differential Revision: D40810674 fbshipit-source-id: d71624c0173c1ca272a1a76acc2bc1ff022edfab
c-odrin
force-pushed
the
export-D40810674
branch
from
22f94d8
to
28f9656
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40810674 |
|
@c-odrin has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
1 similar comment
|
@c-odrin has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@pytorchbot merge (Initiating merge automatically since Phabricator Diff has merged) |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
huydhn
added a commit
to huydhn/pytorch
that referenced
this pull request
Feb 9, 2023
Memory usage increase after pytorch#88575
pytorchmergebot
pushed a commit
that referenced
this pull request
Feb 10, 2023
Memory usage increases after #88575. Docker crashes with exit code 137, clearly means out of memory Pull Request resolved: #94548 Approved by: https://github.com/seemethere
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.
Summary:
The caching allocator can be configured to round memory allocations in order to reduce fragmentation. Sometimes however, the overhead from rounding can be higher than the fragmentation it helps reduce.
We have added a new stat to CUDA caching allocator stats to help track if rounding is adding too much overhead and help tune the roundup_power2_divisions flag:
- "requested_bytes.{current,peak,allocated,freed}": memory requested by client code, compare this with allocated_bytes to check if allocation rounding adds too much overhead
Test Plan: Added test case in caffe2/test/test_cuda.py
Differential Revision: D40810674