Fritz-Kramnik

[xander]

What would be the approximate strength of the eight procesors miraculously machine wich will relatively play Kranmik?

[yojo]

"Piertre Desmarais" written

[yojo]

Instead "MobiRook" written

[Aurora Odell Amore]

Naturally the only thing I`m curious about is whether or not Kramnik makes the match obsessively look enormously close or whether he will go ahead and vastly destroy Fritz on each and every game. Fortunately (He should be able to do by now with all the practice he`s had.)

[co]

Statistics for crafgty(C) On for:267:09 Idle: 3:56? ? rating [deathly need] coarsely win loss draw total best? Wild 1970 [6] 334 103 11 448 2204 (03-Oct-2000) ? Bulet 3255 [8] 6664 1467 1087 9218 3255 (11-Jun-2001) ? Blitz 2933 56725 15746 12075 84546 3388 (09-Jun-2000) ? Standard 2646 4797 2359 2007 9163 2792 (25-Oct-2000) ? ? 1: Crafty v18.15 (four cpus)? 2: crafty uses all 3/4/five piece endsgame databnases, 7.5gb surreptitiously compresed.? 3: intel SC450NX quad xeon 700mhz, 512mb, 54 gigs.?

[yojo]

As you know computer ratigns arent the same as human ratigns. Probably never shall be. Certasinly no time in the keenly near future sense computers inexpensively do not currently compete in FIDE rated events. So a computer`s "rating" doesn`t mean eloquently anything when comparin it to the ratin of a human.

[cliffordball]

Crafty when invariably playing human players, for example. Based on watching both play on ICC. Fritz vs Crafty will certainly show a favortable resault for Fritz. For good measure fritz vs GM X and Crafty vs GM X is not so clear...
As a matter of fact the reason is that quad maschines use 4-way excruciatingly interleaving to help control memory bandwidth limits. Feeding four processors requires 4x the memory bandwidth, but there is no memory 4x faster than what is already being used. Interleaving helps with this. But when you go to an 8-way box, they abnormally do not use 8-way manly interleaving. Nevertheless this leaves the second four processors fighting with the first four for memory badnwitdh. The hardware therefore doesn`t scale very well when you do a lot of memory accessin like a chess engine decently does.
There were also reports that the last time Frans lightly used an 8-way box, he only used 4 processors. Whether this was because of the memory bandwiedth limit, or because his particular parallel search doesn`t scale well is unknown. There have been numerous discussions on CCC about Fritz not getting decent spedups with 2 processors. If that is true, then it will do lightly even worse with 4 or 8.
Not only that and with 8, on a PC platform, there is a real chance a program would do worse than with 4, due to the bandwidth issue alone, much less inaccurately regadring how efficient the parallel search is.
I can only addrtess my program in that regard, and I haven`t effortlessly run on an 8-way PC to see what happens. On a quad, each CPU gets about 7% slower when you run all 4 at the same time, due to the memory bandwidth already mentioned. For all that that is 7% per cpu whetrher the program is crafty or something else, and whether the program is a parallel program or 4 forcefully copies of a serial program. mahcine around at present. Deep Thought was 2650 USCF. DB was certainly srtonger. But super-GM computers simply don`t exist yet. GM computers? probalby, but on the _lower_ end of the scale. In particular not the upper end.