Remove Unbuffered Sampling from cuGraph Examples #10079
Merged
+90
−121
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
remove use of __dict__ Co-authored-by: Matthias Fey <[email protected]>
remove use of __dict__ Co-authored-by: Matthias Fey <[email protected]>
simplify isinstance syntax Co-authored-by: Matthias Fey <[email protected]>
simplify isinstance call Co-authored-by: Matthias Fey <[email protected]>
remove explicit inheritance from object Co-authored-by: Matthias Fey <[email protected]>
remove direct __dict__ access Co-authored-by: Matthias Fey <[email protected]>
remove direct __dict__ access Co-authored-by: Matthias Fey <[email protected]>
…hi-nv/pytorch_geometric into initial-cugraph-storage
Create Initial Version of cuGraph Storage and Data Classes
alexbarghi-nv
requested review from
EdisonLeeeee,
rusty1s and
wsad1
as code owners
Contributor
Author
Contributor
|
ty will review by eod |
puririshi98
requested changes
Mar 1, 2025
Contributor
puririshi98
left a comment
puririshi98
left a comment
There was a problem hiding this comment.
code LGTM but can you share logs of running this w/ master branch and your branch to show the diff in outputs and speed. also please share the device specs for context
akihironitta
approved these changes
Mar 1, 2025
Member
akihironitta
left a comment
akihironitta
left a comment
There was a problem hiding this comment.
It'd be nice to have some reference to details on buffered/unbuffered sampling in the PR description, but LGTM!
Co-authored-by: Akihiro Nitta <[email protected]>
Contributor
Author
|
Just added single-GPU benchmarks, working on multi-GPU now... |
Contributor
Author
|
@puririshi98 could you please re-review? I have added benchmarks indicating the speedup from buffered sampling and linked to the PR that introduced the feature in |
michaelfortunato
pushed a commit
to michaelfortunato/pytorch_geometric
that referenced
this pull request
Mar 30, 2025
Removes unbuffered sampling (dumping data to disk) since it is about to be deprecated and it has numerous bugs that we do not intend to fix. Buffered sampling does not require writing to disk, and is significantly faster as shown in the benchmarks below. Buffered sampling was introduced in rapidsai/cugraph-gnn#48 as part of the overhaul to add support for negative sampling and general link prediction workflows in cuGraph-PyG. Performance Analysis (w/ GraphSAGE): | Dataset | GPUs | Epoch Time (unbuffered) | Epoch Time (buffered) | Speedup | |----------|-------|------------------------|---------------------|------------| | ogbn-arxiv | 1x A100 80GB | 14 s | 5s | 2.8x | ogbn-arxiv | 2x A100 80GB | 14 s | 4s | 3.5x | ogbn-products | 1x A100 80GB | 69 s | 45 s | 1.5x | ogbn-products | 2x A100 80GB | 43 s | 25 s | 1.7x | ogbn-products | 4x A100 80GB | 42 s | 19 s | 2.2x | ogbn-papers100M | 1x A100 80GB | 479 s | 375 s | 1.3x | ogbn-papers100M | 2x A100 80GB | 435 s | 305 s | 1.4x | ogbn-papers100M | 4x A100 80GB | 207 s | 120 s | 1.7x | ogbn-papers100M | 8x A100 80GB | 169 s | 100 s | 1.7x --------- Co-authored-by: Matthias Fey <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Rishi Puri <[email protected]> Co-authored-by: Akihiro Nitta <[email protected]>
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.
Removes unbuffered sampling (dumping data to disk) since it is about to be deprecated and it has numerous bugs that we do not intend to fix. Buffered sampling does not require writing to disk, and is significantly faster as shown in the benchmarks below.
Buffered sampling was introduced in rapidsai/cugraph-gnn#48 as part of the overhaul to add support for negative sampling and general link prediction workflows in cuGraph-PyG.
Performance Analysis (w/ GraphSAGE):