I have done tests on this question in a now rather distant past.
At least with chessprograms from 2007/8 on a single core increasing hashtable size above 16MB didn't really improve the playing strength in a measurable way at all ( as far as I recall with the single exception of Junior).
This was certainly true for Crafty btw where I had a huge enough number of games for the resullts to be statistically significant. At least on a PC Crafty with 16 MB Hashsize wasn't weaker than with 64 or 256 MB Hash.
I got improvements on testpositions with bigger hash sizes but the relevance for game results seemed to be hard to measure if it even existed.
So I am not suprised to read this statement about Toga, but I wonder whether you'd get different results with other engines , say your own one.
Uri Blass wrote:I read a claim that toga unlike other engines does not earn or almost does not earn from more hash tables relative to the default hash of 16 MB
I haven't done any of the tests you mention. But I do run toga on an embedded device on FICS. It has 32MB ram total. I can say that running on 8MB hash is no serious handicap-- people have even accused me of lying about the hardware, so I would say it plays better than most people expect even on just 8MB hash.
Peter Berger wrote:I have done tests on this question in a now rather distant past.
At least with chessprograms from 2007/8 on a single core increasing hashtable size above 16MB didn't really improve the playing strength in a measurable way at all ( as far as I recall with the single exception of Junior).
This was certainly true for Crafty btw where I had a huge enough number of games for the resullts to be statistically significant. At least on a PC Crafty with 16 MB Hashsize wasn't weaker than with 64 or 256 MB Hash.
I got improvements on testpositions with bigger hash sizes but the relevance for game results seemed to be hard to measure if it even existed.
So I am not suprised to read this statement about Toga, but I wonder whether you'd get different results with other engines , say your own one.
Peter Berger wrote:I have done tests on this question in a now rather distant past.
At least with chessprograms from 2007/8 on a single core increasing hashtable size above 16MB didn't really improve the playing strength in a measurable way at all ( as far as I recall with the single exception of Junior).
This was certainly true for Crafty btw where I had a huge enough number of games for the resullts to be statistically significant. At least on a PC Crafty with 16 MB Hashsize wasn't weaker than with 64 or 256 MB Hash.
I got improvements on testpositions with bigger hash sizes but the relevance for game results seemed to be hard to measure if it even existed.
So I am not suprised to read this statement about Toga, but I wonder whether you'd get different results with other engines , say your own one.
