Author: Robert Hyatt
Date: 21:00:07 10/29/99
Go up one level in this thread
On October 29, 1999 at 20:00:57, Bertil Eklund wrote: >On October 29, 1999 at 14:00:20, Robert Hyatt wrote: > >>On October 29, 1999 at 04:09:54, Bertil Eklund wrote: >> >>>On October 28, 1999 at 23:47:19, Robert Hyatt wrote: >>> >>>>On October 28, 1999 at 13:49:18, Tony Hedlund wrote: >>>> >>>>>On October 28, 1999 at 09:31:22, Harald Faber wrote: >>>>> >>>>>>On October 28, 1999 at 08:44:50, Enrique Irazoqui wrote: >>>>>> >>>>>>>>I don't until there is proof. >>>>>>>>A) CM wasn't tested on P90 so there is no comparison and no evidence that Tiger >>>>>>>>is also best on MMX200 or K6-450 as Rebel9 was best on P90 before but wasn't on >>>>>>>>MMX200. So to conclude from the P90 result that Tiger will also be best on >>>>>>>>faster machines is too early and probably wrong. >>>>>>>>B) I don't like all that Tiger-hype since most of the games and results are not >>>>>>>>tournament games. >>>>>>> >>>>>>>?Hype? is propaganda based on smoke and this is not what I am doing. We have >>>>>>>quite outstanding results coming from several sources. Some results at 40/2, >>>>>>>some at 60/30, and so far all seem to indicate that Tiger is the strongest in >>>>>>>comp-comp. Results are facts, no hype. You may or may not like them, but your >>>>>>>choice of the word ?hype? is most unfortunate. >>>>>> >>>>>>You can bury me for the choice of that word if you want to, I don't mind. :-) >>>>>> >>>>>>Show me the tournament results. How many, who played them on which hardware >>>>>>(2PCs?)? >>>>>> >>>>>>>>Concerning TIger/K6-450 vs Prog X/P90 I think it is really nonsense to play such >>>>>>>>a match, no matter if Prog X has an almost safe rating. >>>>>>>>Take a class-a soccer team, let's call it team A, playing versus a team of the >>>>>>>>lowest regional class (call it Z). Team A wins 10-0. Let another team out of the >>>>>>>>class-a play against Z, call it team B. B wins 12-0. Would you say that team B >>>>>>>>is better/stronger than team A? I would never say that until I have more results >>>>>>>>and games within class-a. And you shouldn't say that too. >>>>>>> >>>>>>>SSDF people explained this very many times already. They might do it again in >>>>>>>this thread. >>>>>>> >>>>>>>Enrique >>>>>> >>>>>>I know and I was referring to that most significant argument. If a member >>>>>>doesn't have 2 K6-450 or 1xK6-450+1xMMX200 but only K6-450+P90, better leave >>>>>>playing programs against each other on these 2 machines. >>>>> >>>>>As Enrique wrote, we have explained this so many times before it's sad to go >>>>>down that road again. >>>>> >>>>>It doesn't matter which ELO the opponents to a new entrance have. We can play >>>>>200 Tiger 12 AMD K6-2 450 games against P90-programs, or we can play 200 games >>>>>against AMD K6-2 450 programs. We will get approx. the same ELO. It's in the >>>>>system made by Arpad Elo. >>>>> >>>>>Tony >>>> >>>> >>>>Have you checked this with any statistical measures? I typically find that >>>>given two otherwise equal programs, one advantage of some sort (faster machine, >>>>bigger hash, better book, etc) tends to exaggerate the rating produced by the >>>>Elo system. One example is doubling the cpu speed seems to make a program some >>>>70 points stronger in computer vs computer, but it does _not_ have that effect >>>>in computer vs human games... In other words, the Elo can be somewhat skewed >>>>without it being intentional. >>> >>>Hallo! >>> >>>Any proof for this, except for a gut-feeling!? >>> >>>Bertil SSDF >> >> >>Yes, although not scientifically rigorous enough to really be conclusive. I >>ran an experiment several months ago. I had a GM that wanted to play a bunch >>of games on ICC over several days, getting ready for some tournament he was >>going to. >> >>He would play about 60 blitz games per day. I let him play one day only vs the >>quad xeon, the next day only vs the quad P6/200, and then on the next two days >>I would swap machine after he had played about 1/2 of his games. The final >>result was that there was little difference in how Crafty did using the quad >>xeon vs the quad P6, yet the speed difference is 2x. >> >>These were _all_ 5 5 blitz, although 5 5 (5 mins on clock, 5 sec increment added >>after each move) is not real fast. >> >>By the same token, my scores vs computers between those two platforms are >>much different, with the quad xeon having a decided edge against every opponent >>the two boxes had in common. So against other programs, the 2x speed was very >>important. But at least, against this one GM, the difference was not >>significant. IE I think the xeon had maybe 3% more wins than the quad P6. I >>don't know what that translates into in terms of Elo, but it isn't much. > >Hallo! > >Thank you for the answer! But if you should believe in this yourself, why don´t >you sell your quad qeon and go for a new honeymoon with your wife to a sunny >place, can be a lot of fun!(this is not ironic but I guess you don´t fully >believe in this, as you seems to be number one on your quad) > >At least I lost even faster against my AMD450 vs my P90 whatever program I use! > >Regards Bertil SSDF If I was only playing humans, I might. But I play lots of computers as well, and speed is important. And note that faster hardware _does_ make the program play stronger... just not 70-100 Elo as is often quoted. Otherwise we need a machine 8x faster and current programs would be in the super-GM range, since that would be 3 doublings or +210 to +300 depending on how you think. Do you really think a quad xeon 800 would be that strong? Not me. Strong, definitely. Super-GM? Not a chance.
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.