Author: Christopher R. Dorr
Date: 05:53:58 03/31/99
Go up one level in this thread
On March 30, 1999 at 17:10:20, Bruce Moreland wrote: > >On March 30, 1999 at 16:09:54, Christopher R. Dorr wrote: > >>On March 30, 1999 at 13:57:39, Bruce Moreland wrote: > >>>If you claim that Crafty is stronger than Ferret on equal hardware in head to >>>head games, certainly in 5 0 blitz, you haven't been paying attention. Sorry, >>>but that's true, and there is a tremendous amount of evidence to back that up. >>> >> >>I'm not necessarily saying that. How do you measure superiority, though? This >>isn't as arcane as it seems. I have a 2150 rated friend who has a *big* negative >>record against some local 1900. Looking at head-to-head, you'd say the 1900 was >>the superior player. Looking at their ratings, you would say that the 2150 was >>the superior player. What are we talking about here? > >That is a good question, yes. Your suggestion was that Crafty on equal hardware >may be as good or better than anything else at blitz. The evidence is the >rating of Mofongo in relation to the ratings of the other computer accounts. > >I argue that Mofongo's rating is inflated when compared with other computers on >ICC. The reason is that it plays the opponents and time controls that will >maximize its rating. Other programs don't do this. Again, I'm not trying to be an advocate of Crafty ruling the universe. :) But I believe that there is some good evidence that Crafty may well be this. Not *that* it is, but evidence that supports this. For example Moonshot on ICC is a Crafty running on an ordinary system (K6 250), that *does* play computers, and reasonable time controls, and has a 2750 rating, peaking at over 2900. This is the kind of computer many of us have. And if you want to focus on human-only play, there is also Moonshot, which is 2650ish on a Cyrix 166 (!) which benches out like a real P5/133 or P5/150. All of this evidence taken together, while not conclusive, demonstrates that the claim cannot simply dismissed out of hand. > >I can't argue about comparative performance against humans, that's just way to >random to talk about, in my opinion. But isn't that what it's all about? At least to me, right now, it is. It appears (as I stated in a different post) that we are not necessarily disagreeing, but perhaps talking about different things. To me human-computer performance *is* the keystone. To others it's not important. But I think the existence of ICC with thousands of players, hundreds of computers, and hundreds of thousands of games can help order some of this randomness. It's not exact, but we can get some good 'ballpark' information from the rating, despite the differing formulae, etc. After al, even in the USCF or FIDE, I can choose to play or not play in a tournament because I'll only be playing masters, or won't have to play many 1800+ types...there is some selectivity there as well, but USCF ratings have been recognized as being reflective of strength. > >I don't believe that mine does any worse than Crafty does against humans, >certainly. And it doesn't seem to do any worse than Crafty against other >computers. And it does well against Crafty on equal hardware or even a hardware >disadvantage. No argument here. Your's may well be the best. I'm not arguing that. I'm arguing for clearer definitions, and for a use and discussion of *all* the available evidence. I'm simply asking for a proper dialectic. > >I'm confused as to how someone can say that Crafty may be better, without >getting into some really murky justification. > >And that's just mine. There are plenty of others that do *very* well at blitz. Again, no argument. > >>>>3. Crafty is designed to be SMP. To say it's not fair to state that Crafty on a >>>>Quad Xeon is better than Fritz on a single PIII is the same as saying that it's >>>>not fair that Deep Blue runs on a specially designed computer. If the question >>>>is what's the best, a very valid way of looking at is to measure their >>>>performance on their optimal machine. Crafty's optimal machine is a Quad Xeon or >>>>PIII, while Fritz's is a single PIII 500.On these optimal machines, I honestly >>>>doubt Fritz's superiority. >>> >>>No way. I argue against this on two grounds: >>> >>>1) Crafty has not been crippled by its port to SMP, at least not to my >>>knowledge. It should run just fine on one processor. At very least the SMP >>>part can be compiled out. SMP is an additional thing, not a transforming thing. >>> >> >>And Fritz should run just fine on that Quad Xeon. Is it Bob's fault that Frans >>didn't do any SMP stuff? Multiprocessors aren't as rare as you might think >>anymore. Certainly not more rare right now that the new PIII stuff. On equal >>hardware, I don't know. What hardware. How about my old 486/66 w/ 4 MB RAM? Bet >>you Crafty will run on it. Fritz5? Doubt it. Lots of people still have these >>around. Where do we target? The lowest common denominator? the highest? Perhaps >>an appropriate way to look at this would be a scale of equipment, that reflects >>the fact the Crafty is probably better than Fritz on very low-end and very >>high-end stuff, and Fritz is better in the middle. > >I'm confused as to how you can come up with this. How do you compare the >software element of two micro programs? The most logical thing to do *this >year* is run them on conservative middle- to high-end system. It's ridiculous >to try to compare them on a weird system that has extremely important features >(extra processors) that one can take advantage of and the other can't. You >might as well run Crafty on an Alpha and expect Fritz to use the FX!32 emulator. If we are asking the question "Which program is best on equal hardware?", then you are of course correct. If we are asking the question "Which program (or system, as you say) plays the best chess, as demonstrated by computer-computer testing?", then I believe your assertion is incorrect. They really are different questions. > The Alpha might be faster than the Intel machine, so arguably it would be a >better platform, right? Not Crafty's fault if Fritz isn't compiled to run on >it. > >It would make a lot more sense to use a PIII than an SMP machine, since both >programs would have been targeted at a similar chip, meaning the PII. You'd >expect them both to scale properly. > >Imagine if you benchmarked these programs on various systems. You can't just >produce a benchmark that shows that Crafty is 4x as fast or whatever, and assume >that this tells the whole story and that everyone will draw the proper >conclusions. You have to footnote this, and explain that something important >has been left out of Fritz because Frans has not written it yet. That's only >fair. > >Anyone who has listened to me knows that I have no problem with Crafty, but the >conclusions being made here are just wrong. I don't believe they are wrong. i believe they are answers to a different question than the one you are asking. > >They won't be wrong in a year or two though, although it will probably be a >while before you won't have to mention that you compared a non-SMP program with >an SMP program on an SMP system. I agree. > >bruce Chris
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.