1 core can be better than 6 cores (Stockfish)

[gordonr]

It may be interesting to see list of test positions when 6 cores are better in average and list of test position when 6 cores are not better in average than 1 core with specific version number to see if there is something common to positions when 6 cores do not do better.

Agreed, that would be interesting, though do you mean compared to "1 core" or 2 cores? Carl's video is showing no benefit for more than 2 cores. My tests are also comparing 2 cores vs more than 2 cores.

[Jouni]

I just finished test with Obsidian 15 and Berserk dev. Both engines score much better with 6 cores than 2 cores. So Stockfish has some clever search things to find combinations with less cores. But when did this happen?

[gordonr]

I just finished test with Obsidian 15 and Berserk dev. Both engines score much better with 6 cores than 2 cores. So Stockfish has some clever search things to find combinations with less cores. But when did this happen?

Please post some positions that shows this effect with Stockfish. I'm not seeing it clearly in my tests. I tried some positions from the test set that Hai posted and I'm not seeing it there either but I haven't had time to test all positions yet.

[Jouni]

I tested older SF versions to see when has this behaviour started. Didn't take long: SF16 behaves expected and it gains a lot from additional threads. But SF16.1 don't benefit from more than 2 threads much. It's already better than SF16 and 6 threads with only 2 threads in test positions!

[Uri Blass]

results may be dependent on the test positions people use so it may be interesting to see the specific test positions when 6 cores are not better than 2 cores and the relevant test suite(maybe in part of the positions 6 cores are better than 2 cores and in part of them 2 cores are better.

[gordonr]

I tested older SF versions to see when has this behaviour started. Didn't take long: SF16 behaves expected and it gains a lot from additional threads. But SF16.1 don't benefit from more than 2 threads much. It's already better than SF16 and 6 threads with only 2 threads in test positions!

With all due respect, while yourself and others don't post the test positions, these results aren't convincing. Let others see for themselves with the same test positions. So far, only Hai has posted his test set and I gave up due to cooks, etc. I'm still looking for even just one position where this claim holds true.

[Jouni]

You can see behaviour in HTC suite. 110 position set 1 minute level:
SF16 2 cores 67 - SF16 6 cores 80
SF16.1 2 cores 76 - SF16.1 6 cores 76

[gordonr]

You can see behaviour in HTC suite. 110 position set 1 minute level:
SF16 2 cores 67 - SF16 6 cores 80
SF16.1 2 cores 76 - SF16.1 6 cores 76

For the SF 16.1 results, how many runs was this averaged over? Please post the 110 positions here so I can be sure I'm looking at the same positions, thanks.