Stockfish 12 has arrived!

[MikeGL]

Code: Select all

/tmp/cc3jJ8IN.s:9: Error: file not found: nn-308d71810dff.nnue 

Try to solve the above problem first, maybe some or, if you're lucky, all errors would follow too.

Make sure you have the nnue file in one of your path.
Haven't tried to compile this myself, but the errors being returned to you are clear enough IMO.
download the said nnue: https://tests.stockfishchess.org/nns
Looks like the file nn-308d71810dff.nnue is the first one listed there.

I am having problems to compile the master of https://github.com/official-stockfish/Stockfish both on Linux and MacOSX:

Code: Select all

make stockfish ARCH=x86-64-avx2
...
/tmp/cc3jJ8IN.s: Assembler messages:
/tmp/cc3jJ8IN.s:9: Error: file not found: nn-308d71810dff.nnue
make[1]: *** [/tmp/ccqKQCil.mk:2: /tmp/ccx6iVxC.ltrans0.ltrans.o] Error 1
lto-wrapper: fatal error: make returned 2 exit status
compilation terminated.
/usr/bin/ld: error: lto-wrapper failed
collect2: error: ld returned 1 exit status

I guess something went wrong.. nn? Who can help?

[syzygy]

How does Stockfish 12 with the NNUE perform at chess960 ? Is the NNUE enabled when the UCI_Chess960 parameter is enabled ?

This is a good question to which I don't really have an answer, but my guess is that it does OK at chess960 because the NNUE network doesn't even know yet about castling rights at all. (NNUE is enabled in chess960 mode.)

[syzygy]

Code: Select all

/tmp/cc3jJ8IN.s:9: Error: file not found: nn-308d71810dff.nnue 

Try to solve the above problem first, maybe some or, if you're lucky, all errors would follow too.

Make sure you have the nnue file in one of your path.
Haven't tried to compile this myself, but the errors being returned to you are clear enough IMO.
download the said nnue: https://tests.stockfishchess.org/nns
Looks like the file nn-308d71810dff.nnue is the first one listed there.

I am having problems to compile the master of https://github.com/official-stockfish/Stockfish both on Linux and MacOSX:

Code: Select all

make stockfish ARCH=x86-64-avx2
...
/tmp/cc3jJ8IN.s: Assembler messages:
/tmp/cc3jJ8IN.s:9: Error: file not found: nn-308d71810dff.nnue
make[1]: *** [/tmp/ccqKQCil.mk:2: /tmp/ccx6iVxC.ltrans0.ltrans.o] Error 1
lto-wrapper: fatal error: make returned 2 exit status
compilation terminated.
/usr/bin/ld: error: lto-wrapper failed
collect2: error: ld returned 1 exit status

I guess something went wrong.. nn? Who can help?

Or just type

Code: Select all

make profile-build ARCH=x86-64-avx2

This will first download the network file. Alternatively, type "make net".

[Geonerd]

Can someone with an AMD Bulldozer or Piledriver series CPU please run some benchmarks on the Modern/Popcnt and Base'/x64 compiles?

Popcnt and above use SSSE3 in the NNUE code, which my 6100 series Opteron chips cannot digest. I'm stuck running the slowest executable. The CPUs can be upgraded for cheap, but wonder if the speedup would even be worth the effort.

Thanks!

[fishpov]

Stockfish 12 NNUE cannot find this
[d]r1b1qr1k/2p3pp/4p3/1pb1PpN1/pn3N1P/8/PPP1QPP1/2KR3R w - - 0 1
1. Rd8 !

The last Stockfish 11 060820 with NNUE and (20200814-1035.bin) find Rd8 in less than 1s
Perhaps a regression ?

[Modern Times]

Can someone with an AMD Bulldozer or Piledriver series CPU please run some benchmarks on the Modern/Popcnt and Base'/x64 compiles?

Popcnt and above use SSSE3 in the NNUE code, which my 6100 series Opteron chips cannot digest. I'm stuck running the slowest executable. The CPUs can be upgraded for cheap, but wonder if the speedup would even be worth the effort.

Thanks!

I have an FX8350 Piledriver, will run them later if someone doesn't beat me to it.

[Modern Times]

Can someone with an AMD Bulldozer or Piledriver series CPU please run some benchmarks on the Modern/Popcnt and Base'/x64 compiles?

Popcnt and above use SSSE3 in the NNUE code, which my 6100 series Opteron chips cannot digest. I'm stuck running the slowest executable. The CPUs can be upgraded for cheap, but wonder if the speedup would even be worth the effort.

Thanks!

AMD FX8350 Piledriver, standard bench command, average of 3 runs.

Time taken in ms

3765 modern
3894 ssse
4212 x64

So the base x64 is 12% slower than modern.

[Jouni]

The classic evaluation doesn't make any sense to use. In 60 + 0,6 match engines loses to SF11! REGRESSION .

Code: Select all

                                 
1   Stockfish 11 64 BMI2    +7  +28/=148/-24 51.00%  102.0/200
2   Stockfish 12  classic   -7  +24/=148/-28 49.00%   98.0/200

[OliverBr]

The classic evaluation doesn't make any sense to use. In 60 + 0,6 match engines loses to SF11! REGRESSION .

Code: Select all

                                 
1   Stockfish 11 64 BMI2    +7  +28/=148/-24 51.00%  102.0/200
2   Stockfish 12  classic   -7  +24/=148/-28 49.00%   98.0/200

You need at least 2000 games in order to get a direction, over 5000 games a real ELO number.

[Jouni]

If engine is not leading after 200 games it's not significantly better. BTW latest SF devs before NNUE were +25-30 ELO ahead of SF11 after 200 games always.