Lion wrote:Well, Stockfish is not using Large Pages right?
The 50-60 ELO improvement of H6 over H5 was already including the Large Pages on both right?
So I assume it is fair to say that ASMFish is 30% faster than Stockfish Dev.
Thus the 30% faster may cover the gap ?!
rgds
No, that is wrong. Because in the testing (CCRL, IPON, ...) that produces the strength difference between Houdini and SF you didn't allow Houdini to use Large Pages. If you would activate Large Pages for Houdini, the Elo difference with the vanilla SF would increase.
This is a test just finished. The sample size is not big enough to make conclusions, but some data is better than no data.
ASMFish 22nd May 2017
Houdini 6
game in 3 mins + 2 sec per move
200 games
opening book = 10 plies deep
Hardware = single core Xeon 2697 v2 (3.00 GHz)
Houdini 6 scored:
35 wins
38 loses
127 draws
(I bought the pro version and but ran using the standard version since I was running on single threads in Cutechess and didn't know if I needed to change the NUMA setting. If this is wrong, please tell me)