Ryzen 9 7900X: Speed test of Stockfish compiles

[bnst]

I just did a quick test of the five different compiles from july 13. on my Ryzen cpu.

stockfish_23071306_x64.exe:
1:00.789: 33/64 5.25 44320559 nps: 729088 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_avx2.exe:
29.400: 33/64 5.25 44320559 nps: 1507502 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_bmi2.exe:
29.289: 33/64 5.25 44320559 nps: 1513215 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_modern.exe:
36.661: 33/64 5.25 44320559 nps: 1208929 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_ssse.exe:
36.745: 33/64 5.25 44320559 nps: 1206165 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3

The position I used was an old favourite "r2q1r2/1b2bpkp/p3p1p1/2ppP1P1/7R/1PN1BQR1/1PP2P1P/4K3 w - -"
Running a single thread and using 4GiB of hash and 6 man EGTBs.

Best regards
anst

[RubiChess]

Why don't you test the avx512 compile? It should work on Ryzen 7xxx and would be interesting to see if it is fastest.

Regards, Andreas

[bnst]

I tested all five on the web site: https://abrok.eu/stockfish/
I would like to test the one you mention too so if you have a link it would be nice.

Regards
anst

[Joerg Oster]

I tested all five on the web site: https://abrok.eu/stockfish/
I would like to test the one you mention too so if you have a link it would be nice.

Regards
anst

https://github.com/official-stockfish/S ... 3-f5ab5832

[Chessqueen]

I just did a quick test of the five different compiles from july 13. on my Ryzen cpu.

stockfish_23071306_x64.exe:
1:00.789: 33/64 5.25 44320559 nps: 729088 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_avx2.exe:
29.400: 33/64 5.25 44320559 nps: 1507502 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_bmi2.exe:
29.289: 33/64 5.25 44320559 nps: 1513215 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_modern.exe:
36.661: 33/64 5.25 44320559 nps: 1208929 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_ssse.exe:
36.745: 33/64 5.25 44320559 nps: 1206165 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3

The position I used was an old favourite "r2q1r2/1b2bpkp/p3p1p1/2ppP1P1/7R/1PN1BQR1/1PP2P1P/4K3 w - -"
Running a single thread and using 4GiB of hash and 6 man EGTBs.

Best regards
anst

I would like to see Chess.com testing Stockfish 16 vs Torch, instead of versus Dragon 3.2 https://www.chess.com/de/computer-chess-championship

[CornfedForever]

I just did a quick test of the five different compiles from july 13. on my Ryzen cpu.

stockfish_23071306_x64.exe:
1:00.789: 33/64 5.25 44320559 nps: 729088 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_avx2.exe:
29.400: 33/64 5.25 44320559 nps: 1507502 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_bmi2.exe:
29.289: 33/64 5.25 44320559 nps: 1513215 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_modern.exe:
36.661: 33/64 5.25 44320559 nps: 1208929 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3 a6a5 e3g5 f6e6 e1d2 b7c6 g5e3 e8h8 h7h8 c8h8 d2c1 h8g8 f2f4 e5f4 e3f4 f7f6 f4d2 g6g5 c3d1 e6f7 d2a5 f6f5 d1f2 f7g6 b3b4 g8e8 g3c3 c6g2 (TB: 485)

stockfish_23071306_x64_ssse.exe:
36.745: 33/64 5.25 44320559 nps: 1206165 pv: f3f6 e7f6 g5f6 d8f6 e5f6 g7f6 e3c5 f8e8 h4h7 d5d4 c5d4 e6e5 d4e3 a8c8 h2h3

The position I used was an old favourite "r2q1r2/1b2bpkp/p3p1p1/2ppP1P1/7R/1PN1BQR1/1PP2P1P/4K3 w - -"
Running a single thread and using 4GiB of hash and 6 man EGTBs.

Best regards
anst

I would like to see Chess.com testing Stockfish 16 vs Torch, instead of versus Dragon 3.2 https://www.chess.com/de/computer-chess-championship

Why publicly test...until you think you can overtake SF 16...or the better development version? No need really.
I would have thought the avx2 would have produced more nps than bmi2.

[Modern Times]

Bmi2 and avx2 have always been very close in speed on my machines. Bmi2 slower on early AMD Zen of course

[AndrewGrant]

By running with 6 man TBs, you probably introduced so much noise outside of the engine, that your measurements are a little inaccurate. I would not be surprised if you test again without, you find some variation in BMI2 / AVX2.

[Ras]

I would not be surprised if you test again without, you find some variation in BMI2 / AVX2.

In the Ipman test with SF 14.1 and NNUE, I got 22.324.947 NPS with BMI2 and 22.147.907 NPS with AVX2 on a Ryzen 5700X. 16 threads, no EGTBs used. So basically the same performance.

[CornfedForever]

I would not be surprised if you test again without, you find some variation in BMI2 / AVX2.

In the Ipman test with SF 14.1 and NNUE, I got 22.324.947 NPS with BMI2 and 22.147.907 NPS with AVX2 on a Ryzen 5700X. 16 threads, no EGTBs used. So basically the same performance.

Always been curious about something.
If you used half your cores (4), would you get roughly 11,162,474 and 5.581.237 pm 2 cores? That is, does it scale evenly or is there a percentage drop off with each doubling?