Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Set the reference nps for SFNNv8:L1-2560 #1814

Closed

Conversation

ppigazzini
Copy link
Collaborator

@ppigazzini ppigazzini commented Sep 24, 2023

Fishtest with Stockfish 11 had 1.6Mnps as reference nps and 0.7Mnps as threshold for the slow worker.
Set the new reference nps to 560knps according to the 65% slowdown of SF16 with SFNNv8:L1-2560 wrt SF11 on the ARCH=x86-64-bmi2
Set the new threshold for slow worker to 203knps according to the 71% slowdown of SF16 with SFNNv8:L1-2560i wrt SF11 on the ARCH=x86-64-sse41-popcnt

compiler clang++ 16.0.6

  • arch=bmi2 (Intel Xeon CPU E5-2680 v3)
sf_base =  1579950 +/-   8197 (95%)
sf_test =   555524 +/-   4331 (95%)
diff    = -1024426 +/-   6115 (95%)
speedup = -64.839% +/- 0.387% (95%)
  • arch=sse41-popcnt (Intel core i7 3770k)
sf_base =  1790486 +/-  19610 (95%)
sf_test =   532610 +/-   6143 (95%)
diff    = -1257876 +/-  13960 (95%)
speedup = -70.253% +/- 0.780% (95%)

Raise worker version to 218, also server side.

@ppigazzini ppigazzini added worker update code changes requiring a worker update update code change to account external change (Stockfish, book etc.) labels Sep 24, 2023
Fishtest with Stockfish 11 had 1.6Mnps as reference nps and
0.7Mnps as threshold for the slow worker.
Set the new reference nps to 560knps according to the 65% slowdown
of SF16 with SFNNv8:L1-2560 wrt SF11 on ARCH=x86-64-bmi2
Set the new threshold for slow worker to 203knps according to the 71% slowdown
of SF16 with SFNNv8:L1-2560i wrt SF11 on ARCH=x86-64-sse41-popcnt

compiler clang++ 16.0.6

- arch=bmi2 (Intel Xeon CPU E5-2680 v3)
```
sf_base =  1579950 +/-   8197 (95%)
sf_test =   555524 +/-   4331 (95%)
diff    = -1024426 +/-   6115 (95%)
speedup = -64.839% +/- 0.387% (95%)
```
- arch=sse41-popcnt (Intel core i7 3770k)
```
sf_base =  1790486 +/-  19610 (95%)
sf_test =   532610 +/-   6143 (95%)
diff    = -1257876 +/-  13960 (95%)
speedup = -70.253% +/- 0.780% (95%)
```

Raise worker version to 218, also server side.
@ppigazzini
Copy link
Collaborator Author

This is the computation using SF16 as reference, as asked on Discord. I can update the PR if it's preferable.

Fishtest with Stockfish 16 (#1735) had 1.184Mnps as reference core nps and 0.462Mnps as threshold for the slow worker.
Set the new reference core nps to 556480nps according to the 53% slowdown of SF16 with SFNNv8:L1-2560 wrt SF16 on the ARCH=x86-64-bmi2
Set the new threshold for slow worker to 207knps according to the 55% slowdown of SF16 with SFNNv8:L1-2560i wrt SF16 on the ARCH=x86-64-sse41-popcnt

  • arch=bmi2 (Intel Xeon CPU E5-2680 v3)
sf_base =  1153418 +/-   3947 (95%)
sf_test =   550131 +/-   1011 (95%)
diff    =  -603286 +/-   4596 (95%)
speedup = -52.304% +/- 0.399% (95%)
  • arch=sse41-popcnt (Intel core i7 3770k)
sf_base =  1164055 +/-   7058 (95%)
sf_test =   531466 +/-   2261 (95%)
diff    =  -632589 +/-   5719 (95%)
speedup = -54.344% +/- 0.491% (95%)

@ppigazzini
Copy link
Collaborator Author

superseded by #1900

@ppigazzini ppigazzini closed this Feb 26, 2024
@ppigazzini ppigazzini deleted the nps_reference_SFNNv8 branch May 11, 2024 11:30
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
update code change to account external change (Stockfish, book etc.) worker update code changes requiring a worker update
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant