- Blind mode tutorial
lichess.org
Donate

Stockfish 17.1 is here + quick survey

#21

@AsperinTWINK said in #18:

We are looking forward to Stockfish 20, which will be smarter than the smartest person!

Stockfish 15.1 released december (i think 17) 2022, 16 got released june 30 2023, 16.1 february 2024, 17 got released september 2024, and 17.1 march 30 2025, so on average it takes 6-7 months for each new version to be released, so if we do the math, SF 20 should (approximately) get released...(im typing this as im doing math in my head lol)

September 2025: SF 18

February/march/april 2026: SF18.1

Septemberish 2026 SF 19

march 2027: SF 19.1

AND ON SEPTEMBER OR SO 2027 stuckfish 20 should be released :)

@AsperinTWINK said in #18: > We are looking forward to Stockfish 20, which will be smarter than the smartest person! Stockfish 15.1 released december (i think 17) 2022, 16 got released june 30 2023, 16.1 february 2024, 17 got released september 2024, and 17.1 march 30 2025, so on average it takes 6-7 months for each new version to be released, so if we do the math, SF 20 should (approximately) get released...(im typing this as im doing math in my head lol) September 2025: SF 18 February/march/april 2026: SF18.1 Septemberish 2026 SF 19 march 2027: SF 19.1 AND ON SEPTEMBER OR SO 2027 stuckfish 20 should be released :)
#22

So by late 2027, early 2028 SF 20 should be out :)

So by late 2027, early 2028 SF 20 should be out :)
#24

@AsperinTWINK said in #18:

We are looking forward to Stockfish 20, which will be smarter than the smartest person!

I think Stockfish is and was smarter than the smartest person since the first version (2008), even on mediocre hardware.

@AsperinTWINK said in #18: > We are looking forward to Stockfish 20, which will be smarter than the smartest person! I think Stockfish is and was smarter than the smartest person since the first version (2008), even on mediocre hardware.
#25

I read there is a speedtest function and tried it out of curiosity. Is there a place where to submit the results?
i got 11M with an 8600G and 8.6M on an 8840u, both with DDR5 6000Mhz.
An autoupload function in SF18 would be nice

I read there is a speedtest function and tried it out of curiosity. Is there a place where to submit the results? i got 11M with an 8600G and 8.6M on an 8840u, both with DDR5 6000Mhz. An autoupload function in SF18 would be nice
#26

Speedtest

Razer 16 2023 laptop: 32 threads, i9-13950HX; 32 GB RAM; no large pages: 18.8 M nodes per second

Samsung Galaxy Book 3 Ultra: 20 threads; i9-13900H; 32 GB RAM; no large pages: 17.4 M nodes per second

Speedtest Razer 16 2023 laptop: 32 threads, i9-13950HX; 32 GB RAM; no large pages: 18.8 M nodes per second Samsung Galaxy Book 3 Ultra: 20 threads; i9-13900H; 32 GB RAM; no large pages: 17.4 M nodes per second
#28

I am a sf user, on Linux. sf is packaged up by my distro, MX linux, so I am still on 12, so I at least have the NNUE.

I cannot take the survey, as I have no log-in.

The test feature is nice for repeatable tests. I have made my own condition for comparing settings.

I do use the threads option, and have found that on AMD, the nps scales linearly with threads up to the number of cores, then linearly at a lower slope with additional threads , via hyper-threading, where presumably adding cores is less efficient.

I am a sf user, on Linux. sf is packaged up by my distro, MX linux, so I am still on 12, so I at least have the NNUE. I cannot take the survey, as I have no log-in. The test feature is nice for repeatable tests. I have made my own condition for comparing settings. I do use the threads option, and have found that on AMD, the nps scales linearly with threads up to the number of cores, then linearly at a lower slope with additional threads , via hyper-threading, where presumably adding cores is less efficient.
#29

re sf vs alphazero (lc0?)

https://www.sp-cc.de/

Stockfish is at the top of the ratings list. Crushes lc0 by 100 celo points

Program Celo + - Games Score Av.Op. Draws

1 Stockfish 17.1 250330 : 3854 4 4 15000 69.6% 3705 48.4%
2 Stockfish 17 240906 : 3843 4 4 15000 68.1% 3706 49.0%
3 Torch 3.1 a512 : 3801 4 4 15000 62.5% 3709 48.7%
4 Obsidian 15.09 a512 : 3771 4 4 15000 58.3% 3711 49.7%
[ Lc0 0.32dev BT4-6077500 : 3753 15 15 1000 (NN vs SF15.1 testing)]

re sf vs alphazero (lc0?) https://www.sp-cc.de/ Stockfish is at the top of the ratings list. Crushes lc0 by 100 celo points Program Celo + - Games Score Av.Op. Draws 1 Stockfish 17.1 250330 : 3854 4 4 15000 69.6% 3705 48.4% 2 Stockfish 17 240906 : 3843 4 4 15000 68.1% 3706 49.0% 3 Torch 3.1 a512 : 3801 4 4 15000 62.5% 3709 48.7% 4 Obsidian 15.09 a512 : 3771 4 4 15000 58.3% 3711 49.7% [ Lc0 0.32dev BT4-6077500 : 3753 15 15 1000 (NN vs SF15.1 testing)]
#30

@entropyfoe said in #28:

I am a sf user, on Linux. sf is packaged up by my distro, MX linux, so I am still on 12, so I at least have the NNUE.

I cannot take the survey, as I have no log-in.

The test feature is nice for repeatable tests. I have made my own condition for comparing settings.

I do use the threads option, and have found that on AMD, the nps scales linearly with threads up to the number of cores, then linearly at a lower slope with additional threads , via hyper-threading, where presumably adding cores is less efficient.

You can download and build SF 17 quite easily on linux.. take less than 5 minutes

@entropyfoe said in #28: > I am a sf user, on Linux. sf is packaged up by my distro, MX linux, so I am still on 12, so I at least have the NNUE. > > I cannot take the survey, as I have no log-in. > > The test feature is nice for repeatable tests. I have made my own condition for comparing settings. > > I do use the threads option, and have found that on AMD, the nps scales linearly with threads up to the number of cores, then linearly at a lower slope with additional threads , via hyper-threading, where presumably adding cores is less efficient. You can download and build SF 17 quite easily on linux.. take less than 5 minutes