Repository navigation
cuda: GGML_CUDA_OP_TIMING=1 per-node GPU time breakdown - #323
Open
professorpalmer wants to merge 1 commit into
Open
professorpalmer wants to merge 1 commit into
professorpalmer wants to merge 1 commit into
Conversation
With CUDA graphs off (GGML_CUDA_DISABLE_GRAPHS=1), records an event before every graph node on the main stream, charges the time to the next event to that node (a fused group to its first node), aggregates by op/type/shape and prints the top 30 every 100 graphs. Launch gaps inflate the small ops; the large mat-vecs read true. Diagnostics only; off by default. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 3164035)
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
One of the five pieces of #285, split per bri-prism's request on #317. Single commit on current
prism(eaecb50c7), applies without conflicts. Diagnostics only, off by default; happy to drop it if you would rather not carry it.What it does
GGML_CUDA_OP_TIMING=1(with CUDA graphs off,GGML_CUDA_DISABLE_GRAPHS=1): records an event before every graph node on the main stream, charges the time to the next event to that node (a fused group to its first node), aggregates by op / type / shape and prints the top 30 every 100 graphs through the ggml log (INFO level, so-lv 5inllama-server). Launch gaps inflate the small ops; the large mat-vecs and GEMMs read true.Why it is useful
It is how the per-op cost of a decode step and of a prefill micro-batch was attributed while tuning the 262k serve, and how a recent prefill regression on this card was traced to a demoted buffer rather than a kernel in a few minutes.