Author: Dan Newman
Date: 16:46:35 05/14/99
Go up one level in this thread
On May 13, 1999 at 18:47:07, Dann Corbit wrote: >On May 13, 1999 at 11:52:35, Rajen Gupta wrote: > >>We should all be rejoicing.all the chess computer message boards have been >>concerned with attaining the HOLY GRAIL of a world championship level desktop >>machine which is also relatively affordable. We are now about to get it in the >>next 3-4 years or so(IF ever). REJOICE (By that time of course there will be >>multi processor IA 64 Intel machines) which just might be able to stand upto >>this monster! Have no fear the competition will be just as strong as it is now. >Multiple IA 64 chips (which are down the road) will be about where the DEC 21264 >is today. You can buy a 5 foot tall DEC NT server with 14 of those. I think >that a single Hsu chip will still toast it (if it really achieves 30M NPS). >Let's do some quick back-of-the envelope math: > >Crafty on a PII 300 gets about 160K NPS. We'll call it 200K just to be >generous. It already has 64 bit integer operations, but we'll say the new chip >is twice as fast. We'll also quadruple the bus speed and assume that we can >operate at full tilt. That give us 200K * 8 = 1.6M NPS/chip. With 14 of them >we get 22.4M NPS (assuming 0 SMP loss). We are on the same order of magnitude, >so given a few favorable engineering breaks, we might actually achieve 30M NPS >with a machine like that. Now, consider the cost of this beast verses Hsu's >chip. Ouch. > Not only that, consider how much work is done at each node. The PII 300 at 200 knps gives us 1500 machine cycles/node. This might correspond to about 1500 machine instructions/node. Now it's been said elsewhere (by Bob I think) that Hsu says that the DB eval circuitry does the equivalent work of about 40,000 machine instructions at a node... In order for a general purpose machine to match this, it would have to do at least (40,000 instr/node * 30 Mnps) = 1.2 trillion instructions per second -- which is perhaps 4000 times as fast as the PII 300. So we might match the DB chip with a mere *500* of the new general purpose chips postulated above. As a chess programmer I kind of hope (a little bit) that this venture doesn't succeed, or at least is too expensive for most people, or is vastly delayed. (I'd get one though, to use as a sparring partner for my program if it's only 200 bucks.) Perhaps "formula" competitions between programs will become the norm... -Dan. >Compete? I don't think so. If the chip does become a reality, computer chess >programming will be radically changed for at least a decade. > >What I think the most exciting outcome would be is the possibilities for >explorations. Imagine if one million of these are sold. The data analysis that >could be performed by that many screaming chess monsters boggles the >imagination.
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.