Author: KarinsDad
Date: 17:17:06 01/29/99
Go up one level in this thread
On January 29, 1999 at 19:15:30, Matt Frank wrote: [snip] >>> This still leaves us between 150 -225 rating points >>>shy of World Champion status on a microprocessor. :::Dream time::: Give me a >>>Williamette microporocessor (Intel, circa, 2001-2002 running at 1600 MHZ with 1 >>>gig ram) and I predict that the elo for Hiarcs 7 would be about 2840. Now that >>>will be World champion status, OK? >>> >>>Matt Frank >> >>At 50 to 70 ELO per doubling, you'll need more than 1600 Mhz to get to 2840. >>Probably more like 6400 Mhz or more (if the doubling formula doesn't break down) >>and Deep Blue was running at about 64000 Mhz equivalence (give or take) when it >>lost the first time around. >> >>KarinsDad > >The Williamette chip will run 50% faster than a comparably MHZ pentium chip, and >the 1 gig Ram will provide an increase in hash tables on the magnitude of 4 >doublings i doubling of hash tables = 7 elo), (1) 64 to 128, (2) 128 to 256, (3) >256 to 512 and, (4) 512 to 1 gig = 28 elo. Specifically, Hiarcs 7 on 200 MHZ 64 >ram = 2576: Consequently, a Williamette processor running at 1600 MHZ = 2400 MHZ >pentium 2. Therefore a 12 fold increase represents 2 * 2 * 2 * .5 = 67.5 elo + >67.5 + 67.5 + 33.75 = (240.25 elo) + 28 elo for the increased hash tables = 28 >elo + 268.25 elo + 2576 elo = 2844.25 elo, OK. > >Matt Frank > >Matt Frank Maybe. but if that was the case, why didn't the first version of Deep Blue (running at about 25x faster than the Williamette processor) not crush Kasparov in the first match? And it only won the second match because Kasparov psyched himself out by playing inferior out of opening book type of lines. Deep Blue was based on Deep Thought, so there is no way that the algorithms were inferior. Current programs run at about 300K nodes per second (for the quick evaluation type programs, i.e. less chess knowledge). On the Williamette processor, that would mean a 3.6M nodes per second. Deep Blue ran at 100M nodes in the first match and 200M nodes in the second. This means that the Williamette chip should be about 32x slower or 338 elo (5 x 67.5) less than the Deep Blue for the first version and 405 elo less in the second version. That would put Deep Blue II at 2981 (2576 vs. computers elo + 405). Obviously way too high. I have a feeling that the doubling equation starts to fall apart at the higher speeds. I will remind you of your prediction in 2 ot 3 years when the programs still aren't beating the superGMs at standard times (and you can remind me of mine). KarinsDad :)
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.