Last Update: 23.04.2022
Dear Chess Friends,
And as usually, I'm pleased to announce that,
Another new idea is born: a tour with speed playing strength!
As 2nd tourney name can be called compile/binary competition
Meanwhile, just for comparing...I realized to include and test many Cfish 11 compiles too...I set Contempt 40 for all ....
And as latest testings, SF15 and Sword 14.9a PB compiles are tested as well...
Note also that all of the current testings are played as Ponder OFFCfish and Stockfish 14.1 compile testings are played as NON-NNUEWhere SF15, Sword PB played as NNUE enabled (via default EvalFile)
1. Test with Stockfish 14.1 NON-NNUE (ArenaGUI, 128 MB Hash):
# PLAYER : RATING POINTS GAMES (%) 1 Stockfish 14.1 64-bit POPCNT : 3611.2 527.5 1000 52.8% 2 Stockfish 14.1 64-bit : 3601.2 510.5 1000 51.0% 3 Stockfish 14.1 64-bit AVX2 : 3600.9 510.0 1000 51.0% 4 Stockfish 14.1 64-bit SSSE3 : 3600.0 508.5 1000 50.9% 5 Stockfish 14.1 64-bit BMI2 : 3599.7 508.0 1000 50.8% 6 Stockfish 14.1 32-bit : 3556.8 435.5 1000 43.5%
---------------------------------------------------------------------
2. Test with Cfish 11 (ArenaGUI, 128 MB Hash):
# PLAYER : RATING POINTS GAMES (%) 1 Cfish_11_x86_x64 : 3626.9 414.5 750 55.3% 2 Cfish_11_haswell_bmi2 : 3621.5 408.0 750 54.4% 3 Cfish_11_zen2_avx2 : 3616.5 403.0 750 53.7% 4 Cfish_11_general_x64 : 3616.2 402.5 750 53.7% 5 Cfish_11_zen1_avx2 : 3615.8 402.0 750 53.6% 6 Cfish_11_zen2 : 3611.0 396.5 750 52.9% 7 Cfish_11_x64_N_ELTO_popcnt : 3600.0 384.0 750 51.2% 8 Cfish_11_zen1 : 3599.0 383.0 750 51.1% 9 Cfish_11_x64_N_ELTO_BMI2 : 3591.1 374.0 750 49.9% 10 Cfish_11_x64_N_LTO_BMI2 : 3590.4 373.0 750 49.7% 11 Cfish_11_x64_N_LTO_popcnt : 3588.0 370.5 750 49.4% 12 Cfish_11_x64_N_ELTO : 3583.1 365.0 750 48.7% 13 Cfish_11_x64_N_LTO : 3583.1 365.0 750 48.7% 14 Cfish_11_x32_ELTO : 3553.8 332.0 750 44.3% 15 Cfish_11_x32_LTO : 3538.7 315.5 750 42.1% 16 Cfish_11_x32_E : 3535.4 311.5 750 41.5%
--------------------------------------------------------------------- 3. Test with Sword-PB 14.9a NNUE (Cutechess GUI, 64 MB Hash):
# PLAYER : RATING POINTS GAMES (%) 1 Sword-PB-14.9a_BMI2 : 3788.2 160.5 300 53.5% 2 Sword-PB-14.9a_SSE41_popcnt : 3771.1 152.0 300 50.7% 3 Sword-PB-14.9a_AVX2 : 3768.1 150.5 300 50.2% 4 Sword-PB-14.9a_AVX : 3766.1 149.5 300 49.8% 5 Sword-PB-14.9a_modern : 3766.1 149.5 300 49.8% 6 Sword-PB-14.9a_SSE41 : 3760.1 146.5 300 48.8% 7 Sword-PB-14.9a_SSE3_popcnt : 3750.0 141.5 300 47.2%
--------------------------------------------------------------------
4. Test with Stockfish 15 NNUE (Cutechess GUI, 64 MB Hash):
# PLAYER : RATING POINTS GAMES (%) 1 Stockfish 15 x64 bmi2 : 3788.0 155.5 304 51.2% 2 Stockfish 15 x64 popcnt : 3785.2 154.0 304 50.7% 3 Stockfish 15 x64 avx2 : 3779.7 151.0 304 49.7% 4 Stockfish 15 x64 ssse : 3779.7 151.0 304 49.7% 5 Stockfish 15 x64 : 3775.0 148.5 304 48.8%
As Final Conclusion, Stockfish 15 and Sword PB 14.9a BMI2 compiles managed to win ! Plus Cfish 11 engine as BMI2 performed as 2nd...not so bad as well On other hand, a little bit strange results via Stockfish 14.1 compiles Because on my current PC, usually BMI2, AVX2 compiles are fastest Btw, the overall Stockfish 14.1's draw percentage is little bit high: 83% Where via Cfish11: overall draw perc. is not so high for sure, just 66% Via Sword - PB: overall draw ratio is high 91% due to NNUE is enabled Via Stockfish 15, the overall draw ratio is high too 92% due to NNUE ... Actually no BIG surprise about the draw stats e.g based on older tests.. Download all compile test games (in PGN): Games / Sword-PB / SF15
Best, Sedat