Author: Robert Hyatt
Date: 07:13:00 07/21/00
Go up one level in this thread
On July 21, 2000 at 01:23:20, Tom Kerrigan wrote: >On July 21, 2000 at 00:11:28, Robert Hyatt wrote: > >>Don't keep up with your history? until the early 90's, computer chess was >>dominated by mainframe computers. IBM 360/91's, CDC Cyber 176's and Stars, >>Cray-1's, Amdahl's, DEC machines, etc. >> >>And what I said was that if you dropped a typical computer chess program into >>a cray, no it wouldn't run very well. My first run on a Cray was under 1K >>nodes per second. 2 years later on that same cpu we were doing over 20K nodes >>per second. > >Okay, maybe you were competing against similar hardware. But as you said, you >put in the effort necessary to take advantage of that hardware. Did other >programmers? You seem to imply that they didn't. No... you only interpreted it to be so. Do you _really_ think that Slate didn't work to make his program fast on the Cray. After writing chess 4.0 in assembly for the CDC to make it fast there? Do you know Burt Wendroff and Tony Warnock at Los Alamos? Any idea how much time they put in to develop vectorizable chess code to go fast? Anybody that used a Cray "used" the Cray. You don't just walk up to somebody and say "Can I borrow your 60 million dollar computer to play chess, I have something I threw together over the weekend and I'd like to burn your machine for several nights at a chess tournament... That doesn't happen... > So my original point stands, >i.e., that you had a significant hardware advantage. Consider it this way. Let's >say you have two 4-way PCs and you play Crafty vs. Genius. They're running on >the same hardware, but Crafty is using 4 times as many processors as Genius. >It's a gray area, but I'd still call it a hardware advantage. Never said otherwise. I have a hardware advantage _now_ because I took the time to write a parallel search for Crafty. Rather than spending time on other things like search extensions, or new evaluation terms, I spent the time on the parallel search. Was that time wasted? No. Would it have been better to spend time on something else instead? Impossible to say. But until you do a parallel search, and see just how much work it is to get something that works reliably and correctly, you don't appreciate the fact that it isn't "just a hardware advantage for free". It takes a huge amount of work to get that "free hardware advantage". Same thing on the Cray. You've probably never seen one, much less programmed on one. How do you have _any idea_ of the work needed to make a chess engine go fast on one? > >>You don't know _anything_ it seems. Do you know what SMP means?:: >> >>Symmetric Multi Processing. _not_ message passing on a cluster of machines. >>And parallel Cray Blitz certainly pre-dates zugzwang. Find a parallel program >>that dates back prior to 1982. You will find Ostrich. On a cluster of Data >>General Novas. Need I say more?? > >"You don't know _anything_"?? Christ, don't you realize that you can argue a >point without using inane personal insults? Please look in mirror. Pot. Kettle. I only _responded_... > >And I do know what SMP means, and I stand by what I said. Feldmann has run >Zugzwang on an SMP PC. Not a cluster. He told me so himself. And this was long >before Crafty was SMP. You're in no position to contradict me. Sure I am. Should I tell you when the first SMP PC came along? Of course, as I mentioned, I was SMP long before Feldmann was writing even the beginnings of a chess program... so it doesn't particularly matter. In any case, Zugzwang is _not_ an SMP search program. Never was. Still isn't. It is a message passing program. Just because it might run on a SMP machine doesn't mean a thing as far as shared-memory parallel programming goes... > >Just face it, you didn't invent SMP PC chess programs. If you think I'd believe >that crap, you might as well start telling me that you invented the Internet. I never said I did. I did write the first shared memory parallel search that I know of. In 1982. And I have been doing it ever since. The other parallel search programs I knew of at the time were all cluster-based. So believe what- ever crap you want to believe. Or quote a program that ran on a SMP machine prior to mine? > >>I think everyone would agree that the WCCC is much tougher than the WMCCC. >>Ask any of the micro guys that attended. THe WMCCC was _started_ to encourage >>micro program development as they were not competitive with the "big iron" >>machines. > >So if the WMCCC is so much easier than the WCCC, why don't you win? > I'll win when you grow up. Hopefully sooner... >>History, son, is the way to learn things. >> >>It is there for the reading... > >Do not call me son. > >-Tom Then act like an adult...
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.