pytorch / pytorch/kineto

CUDA time difference between print function and Profiler TensorBoard

Open
#935 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

plugin
Dominant language
C++
Stars
992
Forks
270
Avg merge
2d 11h
Merged PRs (30d)
12

Description

I try to analyse the time spent on CPU and GPU using the print function and Profiler Tensorboard. But I get the different result about the time on CPU and GPU.
For print(prof.key_averages().table(sort_by="cuda_time_total", row_limit=20)), I get the following result:
image
The above result shows that CPU time is more than GPU time. But when I use the tensorboard to show the result, the GPU time is counted much more:
image
I am not sure which of these is the correct result, or am I misunderstanding them?

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No source file, test, or reproducible example is named. Start by comparing the reported profiler table call with the TensorBoard view and determine which CPU and GPU metrics each presents. Done means the differing measurements are explained and, if applicable, the relevant behavior or documentation is corrected.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
observability-sre, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.