My UHO-Top15 Ratinglist is the world's first engine-ratinglist, using UHO-openings, and the world's first ratinglist offering additionally Gamepair-statistics.
NN-testrun of Lc0 0.31dev BT4-6077500 finished. BT4-nets make impressive progress...
https://www.sp-cc.de/nn-vs-sf-testing.htm
(Perhaps you have to clear your browsercache (press STRG+SHIFT+DEL) or reload the website))
SPCC: Testrun of Lc0 6077500 finished
Moderator: Ras
-
pohl4711
- Posts: 2697
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
-
Jouni
- Posts: 3619
- Joined: Wed Mar 08, 2006 8:15 pm
- Full name: Jouni Uski
Re: SPCC: Testrun of Lc0 6077500 finished
"Next testrun: Stockfish 16.1". But you have already tested it as 17.1. version is same.
Jouni
-
pohl4711
- Posts: 2697
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
Re: SPCC: Testrun of Lc0 6077500 finished
Not 100%. When I start SF 240217 avx2 with "go depth 25" and SF 16.1 avx2 with "go depth 25", the number of searched nodes and the pv-line are identical. But the eval differs: SF 240217 shows +0.34 and SF 16.1 shows +0.33, when the calculation is finished. The WDL-calculation differs in SF 240217 and SF 16.1. Why this has an impact on the classical eval? No idea, but there is a small difference.
So, for me it makes sense, to repeat the test with the official SF 16.1 release. And I always tested the official binary from the official Stockfish website in the last years, so I wanted to continue doing this.
And I am doing so many Stockfish-testruns, one more or less does really not matter...