Author: Robert Hyatt
Date: 19:32:58 04/26/01
Go up one level in this thread
On April 26, 2001 at 22:05:28, Vincent Diepeveen wrote: >On April 26, 2001 at 19:58:43, Robert Hyatt wrote: > >>On April 26, 2001 at 18:11:20, Rajen Gupta wrote: >> >>>hi bob: you mentioned in you recent post regarding hsu's chip being capable of >>>doing 1 billion nodes per sec-is this an advanced version of the "chess chips" >>>which were present oin the deep blue?-as far as i remember the deep blue had a >>>number of general purpose powerpc processors and a number of what they termed >>>"chess chips" if this is so > >>NO... here is the math. DB2 used 480 chess processors. About 1/2 of the >>processors ran at 20mhz, the other half ran at 24mhz. 20mhz turns into 2M >>nodes per second, 24 turns into 2.4M nodes per second. The easiest way to >>evaluate this is 480 * 2.2M nodes per second, which is roughly 1 billion nodes >>per second peak. He also reported that he drove the chess chips at about >>70% duty cycle with the SP2. So the actual NPS was 480 * 2.2M * .7, which >>is roughly 700M nodes per second. He scaled this back to 200M, I assume, >>based on the search overhead which makes many of the nodes searched not >>necessary if the search wasn't parallel. IE in Crafty, on a quad xeon/700, >>I search about 1.5M nodes per second peak. But roughly 1/4th of that would >>not be searched by a single cpu, so my "effective" nps is lower. Say 1.1M or >>so. > >>I assume this is why Hsu quoted 200M, because he certainly gave the 480 >>number, and the 2.0/2.4M nps per chip number in several different places... > >200M nodes a second let's get to that. .7 i never read that number >anywhere, woudl be amazing if you can keep 'em busy for .7 considering >a single SP processor addressed 30 processors... You don't understand how it worked. It is trivial to drive them at 100% duty cycle, but then the SP processor would be waiting on the chess processors. It would be a rare case where the speed of the chess processors synced up with the speed of the SP so that _everything_ could run at 100%. > >Anyway the question of Rajen was probably : "why does Hsu think that >if he builds a new cpu it is as powerful as hundreds of chips which >in 1997 shaped deep blue 2". > Yes, but he misunderstood my statement. I wasn't talking about a single chip at 1B nodes per second. That was the peak NPS for the deep blue hardware. 480 processors * 2.2M nodes per second (average of 2.0 and 2.4) per processor. >Well that's easy to answer. The deep blue chips were even in 1997 made >from outdated technology. 0.60 microns. Todays standard is 0.18. Note >GEOFORCE3 graphics card is already 0.15 microns and real soon we'll >see 0.12 or 0.13 whatever getting used. > >Suppose Hsu would redesign those 20Mhz and 24Mhz processors (new info >to me btw) in 0.12 or 0.13 technology, then a single cpu would be >faster as 480 chessprocessors were in 1997. I don't doubt that! I do... I don't think it would run hundreds of times faster. He _might_ drive it to 500mhz to 1ghz... 500mhz would be 25 times faster. 1ghz would be 50 times faster. I doubt it could go faster than that.. > >I have no idea to how many Mhz it would be clocked but it sure would >be way faster! > >Compare for example that we can buy machines of 1 Ghz. I remember >still a few years ago i had a 20Mhz 386sx (16 bits/32 bits) machine. >that's more as a factor 50 difference in speed though in advantage of >my nowadays fast Ghz machine :) > > > >> >> >> >> >>>1)how can one single chip be twice as powerful as the entire deep blue, which >>>had hundreds of processors? >> >>You mis-read what I said. I was talking about the peak NPS for the full DB2 >>machine. 1,000 M nodes per second. >> >> >> >> >>>2)are these chess chips of deep blue and the superchip of hsu good for only >>>chess calculations or can it act as a general purpose processor? >> >>Special chess purpose only... >> >> >> >>> >>>thanks >>> >>>rajen
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.