Author: Robert Hyatt
Date: 10:54:50 06/13/99
Go up one level in this thread
On June 13, 1999 at 12:00:28, Peter Hegger wrote: > >[snip] > > >Hard data: Beat kasparov in 1997 using 480 of thse things. Current version >>of the chip searches at 2-2.4M positions per second per chip. Using an old >>fab process. Commercial version (.18 micron I think) will search at an >>estimated 36M nodes per second per chip. He could possible (for high-end >>users) put maybe 4-8 of these chips on a PCI card. Which would be about as >>fast as the old deep blue hardware. Whether the PCI bus could sustain the >>bandwidth to handle multiple cards is unknown to me... but with search tuning >>it could possible work... > > [snip] > >Bob, as I've been following the DB jr. threads I've seen it mentioned that 4 >chips per PCI card would be the max. If 8 were a reality this would be very >exciting as it would eclipse DB 97 in search speed. >Sorry, I don't have access to IEEE journal but is this where the figure of 8 >possible chips per PCI card was mentioned? >You also mentioned in another thread that DB jr. would be 400 points stronger >than any micro programs of today. Was that assessment based on a 1 chip machine >or the fully blown 8 chip screamer? >Thanks >Peter It was a gross under-estimate. In the IEEE paper, Hsu reported something like a 38-2 result vs micros, using a _single_ 2.4M nodes per second chip, running at about 10% of its normal speed. So the new chip will be on the order of 150 times faster than that chip. Only +400 would be embarassing for that machine... As far as 8 chips per card, that was just a guess on my part... Probably depends on the physical size of the ASIC and support chips, plus current drain limits on the PCI bus, plus bandwith at 100mhz (or maybe 133 by then). If Hsu said 4 in the article, then 4 it is... It will be something. And I personally belive it will happen. If you are a WWF fan, I'll paraphrase Stone Cold Steve Austin: "It will happen because Hsu says so." :)
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.