Für jeden zugänglich im Fishcooking-Forum.
Hier das Posting von Marco:
Zitat:
Ok, I have just sent to Martin the link of the binary for next stage:
http://abrok.eu/stockfish/d34bb889b185e6796bbebb9bb3c2b85c8132d7cb/win64modernsse/stockfish_13110211_x64_modern_sse42.exe
This is the version without TB.
Here is the rationale:
With TB current speed is about 19 Mnps, that is much lower than other engines. I don't remember exactly, but I am not sure we were so slow compared to others with stock version.
People have tested enough TB to be quite confident it can help in matches, but we have never tested with 16 parallel threads, I have given a (very quick btw) look at the code and it seems to me that locking is used to synchronize multiple threads access. Maybe this _could_ be the reason of slowdown (if there is a slowdown).
With TCEC we have a great opportunity to test with very high number of threads and very long TC, of course number of games are too small to be statistical significant, nevertheless some metric, such speed, can be verified reliably.
So I'd like to use this stage as a verification step to see if current TB have this scalability issue, and we can verify immediately, already at first game, so eventually TB authors have the time to fix it and in case, if (and this is a big 'if') we reach the super-final we will use the improved TB again.
So the bottom line is I really wish Stockfish to play the superfinal with proved scalable syzygy tablebases....this is an intermediate step to reach that target , although I know the risk of not reaching the final is very high, but not using such an opportunity to verify/improve TB code would seem a waste to me.
In this decision the developer in me prevailed above the SF fan that I also am.