AB-testrun of Fat Fritz 2 finished.
FF2 has a default Contempt of 0, what I believe is not good for ratinglist-testruns (FF2 is very strong and plays versus weaker opponents in a ratinglist-testrun). Because FF2 is a Stockfish-binary (only the nnue-net is different) and Stockfish has a default Contempt of 24, I decided to test FF2 with Contempt=24, too. So, the result of the FF2-testrun is better comparable with my latest SF-testrun (SF 210131): Same Contempt, same 7 opponents...
https://www.sp-cc.de
(Perhaps you have to clear your browsercache or reload the website)
SPCC: Testrun of Fat Fritz 2 finished
Moderators: hgm, Rebel, chrisw
-
- Posts: 2444
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
-
- Posts: 2444
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
Re: SPCC: Testrun of Fat Fritz 2 finished
Here the single results of FF2 and Stockfish 210131 (latest SF DEV testrun in my ratinglist). Same conditions, same opponents, both with Contempt=+24.
Because FF2 uses a net with double size of normal SF nnue-nets, the nps of FF2 are somewhat lower.
On my AMD Ryzen 12core CPU I got these nps for the chess starting position. After 1 minute of calculation:
singlethread:
FF2: 1.15 MN/s
SF Dev: 1.8 MN/s
(-36% less nps)
22 Threads (11 cores), big Hashtables (8 GB):
FF2: 16 MN/s
SF Dev: 18 MN/s
(-11% less nps)
Because FF2 uses a net with double size of normal SF nnue-nets, the nps of FF2 are somewhat lower.
On my AMD Ryzen 12core CPU I got these nps for the chess starting position. After 1 minute of calculation:
singlethread:
FF2: 1.15 MN/s
SF Dev: 1.8 MN/s
(-36% less nps)
22 Threads (11 cores), big Hashtables (8 GB):
FF2: 16 MN/s
SF Dev: 18 MN/s
(-11% less nps)
Code: Select all
3 Stockfish 210131 avx2 : 3723 7000 (+3924,=3014,- 62), 77.6 %
RubiChess 2.0 avx2 : 1000 (+780,=217,- 3), 88.8 %
Slow Chess 2.5 avx2 : 1000 (+694,=302,- 4), 84.5 %
KomodoDragon 1.0 avx2 : 1000 (+227,=751,- 22), 60.3 %
Ethereal 12.75 avx2 : 1000 (+730,=267,- 3), 86.3 %
Nemorino 6.00 avx2 : 1000 (+629,=369,- 2), 81.3 %
Stockfish 12 200902 : 1000 (+142,=836,- 22), 56.0 %
Houdini 6 pext : 1000 (+722,=272,- 6), 85.8 %
4 Fat Fritz 2 avx2 : 3717 7000 (+3854,=3075,- 71), 77.0 %
RubiChess 2.0 avx2 : 1000 (+761,=239,- 0), 88.0 %
Slow Chess 2.5 avx2 : 1000 (+702,=296,- 2), 85.0 %
KomodoDragon 1.0 avx2 : 1000 (+204,=774,- 22), 59.1 %
Ethereal 12.75 avx2 : 1000 (+704,=295,- 1), 85.2 %
Nemorino 6.00 avx2 : 1000 (+644,=355,- 1), 82.2 %
Stockfish 12 200902 : 1000 (+131,=831,- 38), 54.6 %
Houdini 6 pext : 1000 (+708,=285,- 7), 85.0 %