Author: Robert Hyatt
Date: 14:15:47 01/23/00
Go up one level in this thread
On January 23, 2000 at 13:38:21, Tom Kerrigan wrote: >On January 23, 2000 at 12:54:44, Robert Hyatt wrote: > >>talk to the current commercial chess programmers. What is the probability >>that a new program would sell enough to pay Hsu at least $100-150K per year, > >It may be small, but who cares? I'm selling my chess program. I make several >hundred dollars per month. It's not nearly what Microsoft pays me, but you don't >see me sending back my royalty checks. Your reasoning is pathetic. That is the way to win arguments. No logic, so go to name-calling. Hsu doesn't work at IBM any longer. You _did_ see that? He either has, or will have a full-time job soon, with responsibilities to meet. Where is he going to get the _substantial_ amount of time necessary to completely re-design and rewrite all this stuff? And after it runs like a dog compared to current computer speeds, who is going to buy it? If you make several hundred dollars a month selling your program, that's good. For some of us several hundred dollars a month wouldn't pay for the time. > >And aside from money, there is a fair amount of respect and prestige to be >gained by writing the world's best PC chess program. Read carefully: This would _not_ be the world's best PC program. It would be a PC program with a very good evaluation, but running incredibly slowly compared to other current programs... > >>Message passing? Using an array of 16 chess processors? Using the hardware >>in the chess processors to hold the board position, make/unmake moves, etc? > >So the SP was searching 10 plies, but it didn't know what the position was? >Right. Regarding message passing and multiprocessing, that sounds like a lot of >stuff to remove. Not add. > correct. That was what the chess chip did, if you had read Hsu's article. It kept the chess position, made/unmake moves, etc. Makes perfect sense to do it that way, since it becomes a one-cycle operation rather than hundreds as we do in software. This was done in the _first_ Belle machine in 1978. That was the basis for the final Belle machine in 1980. Which was the basis for the original DT and all the successors... >>I am trying to say, apparently unsuccessfully, that rewriting a big chunk of >>hardware stuff to run in software is a big project. The design decisions would >>be totally different to make it efficient. Not just a "port". A total >>re-write. > >I don't disagree with any of this. Here's what I want to know: Why doesn't FHH >stuff his evaluation function in a PC chess program? He already knows all the >terms and weights to use, so it should be easy. If it's really so great, there's >no reason NOT to do it. SPEED? I gave the Cray Blitz example. 20000:50 was the performance ratio between a machine with vector hardware that we had designed algorithms to use and a PC that was actually faster in scalar integer mathematics. There are at least another three-4 orders of magnitude in the hardware advantage that would be given up. It probably could be the worlds best program, if we required only equal-NPS type matches. Hypothetically anyway. But without that, it would be smart but get tactically busted. > >>Special purpose hardware will _always_ dominate general-purpose stuff in terms >>of speed. Always has, always will. > >Did I ever argue this? How about we stick to the conversation? > >-Tom Been trying. You started the idea "doesn't he have the stuff used to define the chips?" I answered yes, but this would not be usable for writing software to act like the DB chip. Then you asked "doesn't he have some software to use so that he can compare what the chip should do to what it really does?" I responded "yes, probably in bits and pieces, but what good is that for a program and I gave the ValidatePosition() example from crafty. It would be a complete re-design to take the hardware and move it to software. And it would take a lot of time, and lose a factor of 1000 in terms of execution speed over the current machines... maybe more. That would change it from a remarkable machine to "just another slow program."
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.