On Tue, Oct 19, 2010 at 11:29 AM, "Ingo Althöfer" <3-hirn-ver...@gmx.de>wrote:
> Hello Don, > > > > > Concerning chess, the SSDF rating list shows: > > > the best engine for mobile phone has some 2,700-rating (Pocket Fritz 3 > > > on 624 MHz), whereas the best engine (Deep Rybka 3) on "normal" > hardware > > > (x64 2GB Q6600 2,4 GHz) hase a some 3,200 rating points. > > > > > > > A better comparison would be Fritz on the PC vs Fritz on the > > mobile phone. > > It is complicated with Pocket Fritz, because only Pocket Fritz 1 > was based on Fritz. Pocket Fritz 2 was Shredder-based, on Pocket Fritz 3 > is based on Hiarcs. (It is just a policy of ChessBase company to give > all of them the name "Pocket Fritz".) > > Perhaps, I should have written more clearly that from the SSDF list > I took the strongest entry (=Rybka 3) for "normal" hardware and the > strongest entry (=PF 3) for "mobile" hardware. > You made that clear, I was just trying to help refine the estimate of how much you lose on a mobile phone. Your estimate is that you lose 500 ELO going to the mobile phone but it's probably more correct to say that you should "expect" to lose only 300-400 ELO. We can simply deduct 100-200 ELO for the apples/oranges comparison which is unfavorable to the mobile phone. Maybe 100 ELO is closer to the truth. It's not a simple comparison because you cannot simply port one program to a totally different platform and consider the comparison fair. Most of the programs were developed and designed for the PC, not the mobile phone - and this is why I would hedge the bet in favor of less difference. Ideally we would want to compare the best possible implementations for each platform but of course this is a fools errand - we can only estimate and guess. If we assume 400 ELO difference that is approximately equivalent to 6 or 7 doublings - does that sound about right for the mobile phone hardware you are comparing? Memory of course is an issue in Chess too, so I'm not sure that changes the formula very much. Chess programs suffer when the hash table is made smaller but when the hardware is significantly slower it takes longer to saturate the hash tables and this should be the same with Go and Chess. (It may be more important for Go however.) > > > > You get something like 60-70 ELO for each doubling of hardware speed. > > Concerning MCTS, also memory size has to be taken into account... > But according to a recent posting by David Fotland 3,000 playouts per > move already give level 3-kyu for MFoG. Expecting more progress in > "evaluation functions" I believe that before 2014/2015 we will have > dan-level programs with less than 3,000 playouts per move. > > Ingo. > -- > GMX DSL Doppel-Flat ab 19,99 €/mtl.! Jetzt auch mit > gratis Notebook-Flat! http://portal.gmx.net/de/go/dsl > _______________________________________________ > Computer-go mailing list > Computer-go@dvandva.org > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >
_______________________________________________ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
