Stockfish builds

2020-09-23 20:11:45
The A.R.B Chess System - 200 + Amazing Wins! By A.R.B :)

A Beautiful A.R.B.C.S Black Win! :)

2020-09-23 19:28:43
yes x64 uses sse2
2020-09-23 19:12:54
I just worked it out! :)

hww and sad13 are the same person doing multiple Posts! :(

How can we STOP! this nonsense!!! ?
2020-09-23 14:51:49
Thanks for your efforts Mrert, always answering everybody's questions...
2020-09-23 09:57:16
Many thanks! :)

Do you happen to know what level of SIMD the x64 code uses? I think it's SSE2, which has been around since the dinosaurs.
2020-09-23 06:41:24
it would be interesting to know how much faster the most modern instruction set are (AVX512, vnni256, vnni512). But you need a very recent high end intel...

Actually x64 is not that slow compared to the first NNUE compiles where it was abysmal, several times slower than bmi2. They did a pretty good optimization job
2020-09-23 06:37:53
Great idea actually.
Running SF12 with "bench 64 1 14 default depth NNUE" (1 thread, depth 14, use NNUE) 3 times per compile:

x64: 654 kN/s
ssse3: 806 kN/s -> 23% faster
modern: 816 kN/s -> 25% faster
AVX2: 940 kN/s -> 44% faster
bmi2: 989 kN/s -> 51% faster

Interesting. modern doesn't bring much more than ssse3, while leaps between other compiles are clear.
2020-09-23 05:45:02
Mrert, could you do us a favor and post benchmarks for the SSSE3 and AVX2 compiles?
My old phenom based opteron system is limited to the pokey x64 executable. I'm contemplating getting a few of the piledriver based 63xx chips since they can at least run the SSSE3 path. I know there are differences in code execution speed between families of CPU, but a bench should give me some idea what to expect. Thanks.
2020-09-23 03:48:15
@Franky: no worries ! I think it's a very interesting issue.
FYI i have a i5-7200u (notebook). On 2 threads, on a random balanced position (with a lot of material), with SF 12 official complies i get :
bmi2 : 1.55 MN/s
modern : 1.2 MN/s
x64 : 900 kN/s

Thanks to the magnificent climate of where I live I currently have no issue with overheating
2020-09-22 21:10:09
sad13 thanks for your expert analysis
2020-09-22 19:01:34
I am not entirely sure but I remember that stockfish had hardcoded endgame knowledge

into it and then the stockfish developers decided to remove this endgame

knowledge as stockfish will depend on tablebases for it.

Maybe stockfish 10 still has this "knowledge" build into it.

I get high eval as well so it must be normal.
2020-09-22 17:23:52
I didn't use tablebases.
But in KNNK ending there is no need. For example, Stockfish 10 running from console immediatly sees that it is draw.

Stockfish 10 64 POPCNT by T. Romstad, M. Costalba, J. Kiiski, G. Linscott
position fen 8/3k4/8/8/8/1N6/2KN4/8 w - - 0 1
go depth 30
info depth 1 seldepth 1 multipv 1 score cp 0 nodes 17 nps 17000 tbhits 0 time 1 pv b3a1
info depth 2 seldepth 2 multipv 1 score cp 0 nodes 128 nps 64000 tbhits 0 time 2 pv b3a1 d7c6

info depth 30 seldepth 30 multipv 1 score cp 0 nodes 4587023 nps 3927245 hashfull 93 tbhits 0 time 1168 pv b3a1 d7c6 a1b3 c6b5 b3c1 b5a4 d2b3 a4b4 c2d2 b4b5 b3a1 b5a4 d2e2 a4a5 a1b3 a5b6 e2f2 b6b5 f2f1 b5b4 f1e2 b4a4 c1a2 a4b5 a2c1 b5c6 e2e1 c6d6 e1f2 d6e6
2020-09-22 10:31:33
>not know if losing position but ofc every loss can be avoided if you play the right moves early enough :P

If the position is losing there are no right moves.

All the moves are losing. If they weren't then it wouldn't be a losing position.

>slight white advantage out of book?

It was a huge advantage, maybe winning for white.

>disapoint loss

I think it's almost certain that stockfish will win the reverse.

That opening gives a huge advantage to white.
2020-09-22 10:29:02

Are tablebases installed and used?
2020-09-22 01:43:15
16...Ne5 17.Qe1 (+1.55 42/62 ply, tb 1k SF 210920) instead of 16...g6 could have held against Komodo.
2020-09-22 00:44:39
@Mrert,no constant pace..sometimes slow,sometimes fast..never mind it's not such a big deal..I'll just scroll down the list and pick a less advanced binary...
2020-09-22 00:09:06
the misguided idea of embedding is not suited to frequent downloading of abrok binaries as you will wastefully download the same net again and again multiple times

repeatedly wasting storage and wasting bandwidth

the links of the nnue nets have been removed from the download page because of embedding

roman can fix this
31034 @roman set up poll so he know what we the users want

result has to be clear enough that he remove embedding from his exes here

please restore sanity to stockfish

Vote default net NOT embedded
2020-09-21 21:27:54
’ was it avoidable though? ’

not know if losing position but ofc every loss can be avoided if you play the right moves early enough :P

slight white advantage out of book?
but 18. Nxe6! 10 moves after last book move
speculate plenty of time?
i not confirm with my hardware
maybe another net would have been better
much depends on net now
disapoint loss
latest patch affect nnue again
better or worse?
2020-09-21 21:15:12
Stockfish thinks that white has winning advantage in KNNK ending.
2020-09-21 20:58:30
was it avoidable though? Maybe the position was already losing...

It's a bit sad that stockfish didn't hold, it seems like it would have been

a fantastic hold considering the opening.
2020-09-21 20:15:38
k v sf
2020-09-21 20:14:07
18. Nxe6! put stockfish in trouble
2020-09-21 20:04:39
Still, it would be very cool if stockfish could hold or maybe win with black.
2020-09-21 20:04:17
It was winning right from the opening(or close to winning)

stockfish evaluated the position out of the opening as 1.42

The reverse is going to be interesting.

It has very good prospects of being the shortest game.
31419Lalit Mohan Bankoti
2020-09-21 19:54:15
STOCKFISH lost against Komodo noooooo.

next page >