SPCC: Testrun of Stockfish 200810 finished

[pohl4711]

Testrun of Stockfish 200810 avx2 with new "default"-Framework nnue-net 112bb1c8cdb5 (= sv200810_0744) finished.

https://www.sp-cc.de

(Perhaps you have to clear your browsercache or reload the website)

[mehmet123]

Hello Stefan, can we see NNUE MEA testings? Maybe this will allow us to learn which network is better at problem solving.

[pohl4711]

Hello Stefan, can we see NNUE MEA testings? Maybe this will allow us to learn which network is better at problem solving.

Of course not. That would make no sense. The lc0 nets are much more complex than the nnue nets. They are around 500 times slower in computation. A MEA test of nnue nets with 1 node per testposition - like in my MEA tests of the lc0 nets - would make no sense. And to do MEA tests with nnue nets with a small thinking-time instead, would make the results uncomparable to the MEA tests of lc0 nets.

Thats why I do my 500 games quicktests of nnue-nets (60''+500ms) vs. Stockfish 200717.

Both methods have the same goal: to test a lot of nets really quickly and find the best nets for the "real" testruns. But it makes no sense at all to use one method for both kind of nets. Because nnue and lc0 nets are so extremly different.

[Leo]

Testrun of Stockfish 200810 avx2 with new "default"-Framework nnue-net 112bb1c8cdb5 (= sv200810_0744) finished.

https://www.sp-cc.de

(Perhaps you have to clear your browsercache or reload the website)

A new record for computer chess Elo.