McCain v10 Release

[MikeB]

Notes and all flavors of exe's can be found here:

https://github.com/MichaelB7/Stockfish/releases/tag/V10

Thanks,
Michael

[MikeB]

Fro those who want to see how their machine stacks up against an iMac (2015) model:

Code: Select all

iMac-3:src $ McCain-v10_x64_bmi2-gcc_mb b >/dev/null

===========================
Total time (ms) : 1468
Nodes searched  : 4063570
Nodes/second    : 2768099

and here's a Raspberry Pi 3B+ (not overclock):

xxxxxxx@PicoChess:~/McCain-v10_update-v2/src $ ./McCain-v10 b >/dev/null

===========================
Total time (ms) : 14865
Nodes searched  : 4063570
Nodes/second    : 273364

Please feel free to post your bench results in this thread.

[peter]

Notes and all flavors of exe's can be found here:

https://github.com/MichaelB7/Stockfish/releases/tag/V10

Thanks a lot, Michael!
None of the MF-McCain- compiles work at Fritz14, maybe just because of the long names again, but as for Shredder, which doesn't care for name- lengthes normally, it's the same.
Trying MF-McCain compiled by you, Shredder(13) gives error- message "cygstdc++6.dll is missing at you PC" (might be not true, but don't know, where to look for it), Stanback- compile: "this is not an UCI- engine".
Lucs-Monge- compile works at Shredder and Fritz.

[Norbert Raimund Leisner]

Hello Peter,

cf. these websites

https://www.solvusoft.com/en/files/miss ... tdc-6-dll/

https://www.dllme.com/dll/files/cygstdc_-6_dll.html

https://de.azdll.net/files/cygstdc-6-dll.html , just click on herunterladen (=download) and confirm before that you are not a robot etc.

Norbert

[mar]

Waiting for McLeela

[BrendanJNorman]

Waiting for McLeela

Aren't you the author of Cheng?

Your engine is fantastic man, why not incorporate your own ideas with the MCTS/NN model?

Cheng already has a nice strategic style, the combination of different methods would be beautiful.

Not regarding playing strength necessarily, but just the type of chess it might produce.

[mar]

The obvious "idea" would be to create a Leela-SF hybrid, the tricky part being how to combine them.
I can imagine Leela doesn't need all the CPU cores so the AB engine might check for blunders and perhaps take
over in endgames, while Leela excels at openings/midgame.
Also whenever AB engine finds a certain win, it should take over as well.

[peter]

cf. these websites

Thank you, Norbert!

[Spliffjiffer]

1. thx MikeB
2. same prob here as Peter has...only Monge compliles work and meanwhile TB's access seems crippled or rewritten eg: 4K3/4p3/4k3/8/Bp6/6b1/2P5/8 b - - 1 4
(normally 1000's of hits...Monges's compile almost none)...looks strange to me so far

[MikeB]

1. thx MikeB
2. same prob here as Peter has...only Monge compliles work and meanwhile TB's access seems crippled or rewritten eg: 4K3/4p3/4k3/8/Bp6/6b1/2P5/8 b - - 1 4
(normally 1000's of hits...Monges's compile almost none)...looks strange to me so far

I did make a probing change - when you have 7 men on HDD drives , you have to curtail the probing or else it slows to crawl - with nps. less than 100K. I believe the changes I made limit probing to the the 7 men Tb probing with 7 or more pieces - with the 7 men Tb limited to nearly one probe. With more than 7 men , it will probe the 6 men tbs as before. I assuming you are using 6 men tbs and and you are seeing probing limited with exactly 7 men as you posted and with 8 or men on the board , probing is leas than before. Can you check the number of probes with 6 men left of the board , 7 men left on the board and 8 men left on the board? 6 should be normal , 7 will be just a few and 8 will be more than 7 but less than what 8 did before.

With it being being nearly impossible for most people to afford nearly 20 gbs in SSD drives there are not too many good alternatives if you want to use 7 men Tbs . What would be really good if you provide examples where the new code does not find the solution vs the old code.