Author: Robert Hyatt
Date: 11:58:54 01/18/01
Go up one level in this thread
On January 18, 2001 at 13:47:47, Bruce Moreland wrote: >On January 17, 2001 at 15:43:22, Robert Hyatt wrote: > >>And the difference between HiTech and today's programs is _not_ 300-400 >>elo. It is not even 200. Hitech Reached beyond 2450 USCF, which is certainly >>beyond 2400 FIDE. I don't believe we have a 2600 FIDE program around today, >>no matter what the hardware, with the possible exception of Deep Blue itself. > >I think it's very possible that Hitech could consistently lose 3 out of 4 >against modern stuff running on the best hardware, which is what is required for >a 200 point delta. > >I don't think that you make a very strong argument when you take a national >rating against humans, convert it to a FIDE rating, estimate the FIDE ratings of >current computers against humans, then compare. Every aspect of this is >controversial. You could start a 50-post thread about how to convert from USCF >to FIDE, for instance, not that I have any intention of starting it. That really wasn't the central point of my argument. The basic facts about hitech are (a) about 150K nodes per second. Not bad by today's standards, but still low; (b) a reasonable evaluation even by today's standards, because of the way their hardware could recognize things in parallel. So the only thing lacking to compete with today's computers would be speed. I don't think anyone today is good enough to win 3 of every 4 against that machine, unless we take _real_ upper-limit hardware like a 16 cpu alpha. Most programs are going to be 2-3 times faster. That might get 100 Elo on a good day. HiTech's search was not simple nor prone to tactical errors due to their pretty effective search extensions (SE, plus all the classics). IE I would _certainly_ not play Cray Blitz at 200K nps, vs Crafty at 1M nps, and expect to win 3 of every 4 games. It just wouldn't happen. Crafty at 1M ought to win a long match, but no chance for 25-75 I don't think. I think HiTech would produce a significant number of draws against anybody, making 3/4 nearly impossible to reach. > >Computer chess people have always had two problems to solve -- computer vs >human, and computer vs computer. I think that they are very distinct problems. >People who used to use supercomputers to play against 80286 machines were not >receiving much of a challenge, and who knows what kind of bizarre thinking was >adopted without there being much difference in outcome. Everything you do looks >like genius when your opponent is searching 5 plies. Note that they looked pretty strong against IM and GM players as well, as proven by OTB results and their official USCF rating. > >I think the disadvantage of having to pioneer, combined with the lack of readily >available serious computer competition, is such that I doubt that Hitech would >do very well against modern programs running on superior hardware. > >The micro programmers have travelled in a competitive herd for years. Everyone >knows if they are weak or strong. If they are weak, there is great incentive to >make changes in order to get strong -- despair because someone else outsearches >you by a ply on the same hardware is a powerful motivator. This has to result >in stronger computer vs computer play than working in a near-vacuum for years. > >bruce That impetus was always there. HiTech had to worry about Belle (just as fast), ChipTest/Deep Thought (considerably faster), Cray Blitz (considerably faster), to name three. I'm not a big fan of HiTech, for sure. But it could play chess. It would definitely be interesting to see it play vs several of todays programs, over a substantial match duration. If I knew of anyone with an old Cray YMP, I would be interested in cranking up Cray Blitz at 400K nodes per second (this is about a 1987 speed, roughly) to see how it would do. Because it and HiTech were very evenly matched. However that is also unlikely to happen. YMPs are long gone... However, let me remind everyone of one interesting thing. Ken Thompson won the 1980 WCCC event with the new Belle machine. A couple of us caught up with him and then passed him in speed (CB won the WCCC in 1983 and 1986). In 1986 after winning the WCCC we had to pass on the ACM event that year as we couldn't get machine time for two events about 3 months apart. HiTech also passed after losing to us in Cologne. Monty appealed to Ken who brought Belle out of retirement and guess what... Belle won the 1986 ACM event quite handily. The point? Old programs are just "old programs". They are not necessarily "push-overs" by any stretch of the imagination. HiTech had a lot of 'stuff' in it. It could play evenly with the original chiptest/deep thought, and I would bet that nobody here is going to drub the old original deep thought by a 3:1 margin. The machines were tough... Yes I think they might lose a match. But not by a lopsided 15-5 result...
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.