There was a 4~5% variance in the capture results after using the "verbose" parameter.

intel / pti-gpu

Profiling Tools Interfaces for GPU (PTI for GPU) is a set of Getting Started Documentation and Tools Library to start performance analysis on Intel(R) Processor Graphics easily

MIT License

202 stars 57 forks source link

There was a 4~5% variance in the capture results after using the "verbose" parameter. #55

Closed ZePan110 closed 8 months ago

ZePan110 commented 10 months ago

Snipaste_2024-01-15_11-01-03

with "verbose" log: https://osgc-sh-earth1.sh.intel.com/xebot/log/PVC/tensorflow/00049489_aicontainer/hotfix_agama-ci-devel-736.9/AI.TF.Bert_large.Inference.Fp16.bs64.s.Debug/0_AI.TF.Bert_large.Inference.Fp16.bs64.s.Debug.txt with no "verbose" log: https://osgc-sh-earth1.sh.intel.com/xebot/log/PVC/tensorflow/00049427_aicontainer/hotfix_agama-ci-devel-736.9/AI.TF.Bert_large.Inference.Fp16.bs64.s.Debug/0_AI.TF.Bert_large.Inference.Fp16.bs64.s.Debug.txt

ZePan110 commented 10 months ago

test command: with "verbose" log: https://osgc-sh-earth1.sh.intel.com/xebot/log/PVC/tensorflow/00049489_aicontainer/hotfix_agama-ci-devel-736.9/AI.TF.Bert_large.Inference.Fp16.bs64.s.Debug/0_AI.TF.Bert_large.Inference.Fp16.bs64.s.Debug.sh with no "verbose" log: https://osgc-sh-earth1.sh.intel.com/xebot/log/PVC/tensorflow/00049427_aicontainer/hotfix_agama-ci-devel-736.9/AI.TF.Bert_large.Inference.Fp16.bs64.s.Debug/0_AI.TF.Bert_large.Inference.Fp16.bs64.s.Debug.sh

Sarbojit2019 commented 10 months ago

With 'Verbose' option on it is expected to loose the performance. It is hard to draw a line though. In your data/observation it looks other way (Verbose is performing better) which seems to be run to run variation.

Sarbojit2019 commented 8 months ago

@ZePan110, Can you please close this ticket?

ZePan110 commented 8 months ago