Author: Vincent Diepeveen
Date: 07:58:09 05/25/02
Go up one level in this thread
On May 24, 2002 at 12:21:59, Jorge Pichard wrote: >On May 24, 2002 at 11:03:16, Vincent Diepeveen wrote: > >>On May 24, 2002 at 10:04:25, Chris Carson wrote: >> >>an XP1900 from rebel could be faster than a slow quad where >>it's going to play on i guess. >> >>Perhaps it even goes play on a cheap dual K7 which is way >>faster than any of the mentionned machines. You are on drugs, we were discussing processor speeds. Fastest 8 processor is like $100k and uses 900Xeons x 8. No way to get such a machine easily without major sponsor. They couldn't find a big sponsor yet, i see Einstein is organizing match now. Keene is gone. So big sponsor problems here. Fastest 4 processor is 4 x 900 Xeons = 3.6Ghz. Speedup from fritz at 4 processors i would estimate a bit lower than crafty, so being like 2.6 to 2.9 in the difficult middlegames (don't compare it with solving testsets please, the problem of the crafty splitting lies in the hard positions where score drops slowly). Even if it's 3.0 then still 3 x 0.9 = 2.7Ghz of a P3 Xeon. This where K7 is faster for Fritz and also 2 x 1.73Ghz = 3.43Ghz and speedup at 2 CPUs is way better relatively seen than at 4. Even (1.7/2) x 3.43 = 2.91Ghz In short whatever calculation you make you speak about being way faster on a dual K7 when compared to a quad Xeon. Assuming 8 processor machine is too expensive to buy, getting a dual is the cheapest and fastest option (note that speedup at 8 processors will be horrible for fritz). Doesn't take away that for the average idiot who reads reports 4 processors seem faster than 2 in his imagination, mentionning the speeds of the processors isn't interesting even then. I doubt chessbase will show up with a 900Mhz quad also. This processor is pretty expensive to buy. From a sponsor start of this year they got a dual K7 1.6MP, so perhaps they'll go for that. Obviously showing up with a quad means simply they go for the PR, not for the speed. >Can you prove it?, provide us with the results of any match that you have seen >or have done lately. >> >>where do you get a $100k quad from? > >Who ever mentioned a Quad on any of the previous posts? > >> >>>On May 24, 2002 at 08:21:17, Jorge Pichard wrote: >>> >>>>Fritz performance was NOT greater than 2650 and I doubt that if they would >>>>paired it against GM Smirin it would have won at least one game. Deep fritz will >>>>be using an 8X system against GM Kramnik, that has a difference of more than 200 >>>>rating points over GM Gulko and is a better player than GM Smirin. >>>>I doubt if Deep Fritz will gain more than 150 raring points by using an 8x >>>>system at standard time control. Also those games that GM Gulko played were at >>>>G/60 plus 10 seconds per move. I just wonder why Chessbase is scare to match >>>>their best programs at standard time control, like Mr. Schoeder with Rebel. >>>>Chessbase doesn't even want to match their best programs at G/90 plus 5 seconds >>>>per move, which to my judgement GM Gulko , GM Seirawan, or GM Smirin would >>>>demonstrate that the programs are NOT yet over 2600 against them, even if they >>>>use the latest Intel Pentium IV 2.53 Ghz. >>>> >>>>Pichard. >>> >>>Ofcourse you can think what you like. >>> >>>I expect the 8x machine that D.Fritz plays on will give a 2650 performance. I >>>think DF or DJ is 2750 (tournament, not match) on the 8x box. This match will >>>be against a 2800 player that is well prepared. 100 points seems to be the loss >>>for a computer (on average) when paired against a well prepared GM in match >>>play. >>> >>>On single proc AMD or Intel boxes, Fritz, Junior, Tiger, Hiarcs and Rebel are >>>2650 (actually 2680 when results from 800Mhz and faster machine results are >>>calculated). Rebel had a 2700 performance against a well prepared (100 games of >>>preparation) 2700 GM using an AMD 1900+. These are Standard FIDE and 40/2 >>>results. >>> >>>Yes, you will see some performances above and below (200 - 400 points) the >>>average ELO for the programs, but the same is true for the GM's, check TWIC. >>> >>>GM's (strong GM's) have to play flawless chess with white pieces and use >>>anti-computer (closed positions) to defeat the computers. Typical or average >>>GM's get beat by the programs on a regular basis. >>> >>>Enjoy this time. Today, GM's below 2600 have a very hard time with the >>>computers, in 3 years it will be GM's below 2700 on single processor boxex. In >>>3 years on the 8x box, well, only the 2800 GM's will have a chance. Besides the >>>speed improvement, there is impressive sw improvement. In 5 years (+ or - 2 >>>years) the machines will rule this game. They will still lose every once in a >>>while, but it will be very rare.
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.