Author: Vincent Diepeveen
Date: 07:32:48 08/07/02
Go up one level in this thread
On August 07, 2002 at 10:13:45, Jonas Cohonas wrote: i never understand all the nonsense posted here. first of all there are no shared memory 8 processor xeon machines. secondly the dual quad xeon machines are 700Mhz. 4 x 700Mhz = 2.8Ghz add to that that for deepjunior7 here the K7 MP processor is more than 20% faster than a P3-xeon at the same speed. in total we talk about a dual 1.2K7 being way way faster for plydepth. We didn't adress speedup yet. If you search with a few million nodes a second like they do at just 2 k7 or P4 processors, then obviously it is hard to get a good speedup. Way harder than it is for DIEP to achieve. In short out of 4 processors you get real bad speedup. Of course from commercial viewpoint running on a 8 processor box sounds great, but only 4 processors will get used from it, because otherwise they have NUMA problems. Only very slow and old 8 processor boxes are possible to use with 8 processors. those are slower than a single cpu K7 however. For real speedup what they need is the latest experimental P4 core from intel (not that i believe it's faster than a k7 until i have seen proof despite that this intel guy said at the world champs, but he only can talk positive anyway). However please let's be very clear. a 8 processor box *never* is going to outgun a dual machine unless it's having the same processors like are in the dual machine, which is NOT the case. the P4 only runs dual at most. Not quad. not at 8 processors. at the experimental quad p4 boxes, you can only run dual in fact. It's like the 1024 processor box in amsterdam. Only 512 processors can get used at the same time, because they belong to the same partition. So in www.top500.org you will see at place #123 the machine from amsterdam which i used. Yet they quote 512 processors there because that's the maximum you can use for having shared memory. It works like that with 8 processor boxes too. it's 2 quad xeons which are plugged in each other. Crafty's practical speedup at a quad xeon is about 2.8 to 2.9 after a minute 1 to 2 minutes. Of course better and what bob says the 3.1 he gets sometimes after a minute or 10-15. without nullmove it's 3.2 - 3.3 Junior is supposed to do without nullmove. well whatever. but they forward prune. same problem as nullmove. speedup will be 2.8 to 2.9 too. the search depth they get it's harder to get a good speedup simply than the old researches without nullmove/forward pruning. what's 700Mhz Xeon x 2.8 ? right very little. then reduce that by another 20% to compensate for K7. Whatever math you do. it's horrible! >On August 07, 2002 at 10:06:36, Vincent Diepeveen wrote: > >>On August 07, 2002 at 09:07:15, Rex wrote: >> >>>I will be VERY upset if Kasparov "Pay" Chess will be the ONLY site to watch this >>>spectacular match. ICC and a free chess server like FICS should also be given >>>the oportunity to relay the game live. >> >>it is logical of course if kasparochess owns the rights for online play. >>it is very dumb for sales of junior, but i understand it if they negotiated >>this. >> >>the more PR the better of course for a program. >> >>>Not only will this be a huge match but CAN and WILL promote the game of CHESS >>>and should_not_be_limited to only one chess site. There should be MANY outlets >>>on the net to view this match allowing people not familiar with computer chess >>>and chess in general to view this match. >>> >>>ALSO WHAT HARDWARE!!! >> >>Obviously this is easy to guess. the fastest machine for both fritz and >>junior is a dual machine and a dual k7 to be precise. >> >>However, junior gets sponsored by intel a lot. intel obviously is capable >>of providing a reasonable fast dual too (compared to your single cpu machines) >> >>speed of the hardware will not be the decisive factor in this match. >>only whether kasparov wants to show something. look he doesn't get paid >>to win. only 1/10 of th etota price money more. >> >>let's hope he wants to compete with kramnik though and beat junior harder >>than kramnik wants to beat fritz. >> >>knowing kasparov it will be again a great show of the dumb. playing into >>tactical lines and especially taking care he looks good on the photo >>instead of in the analysis of the GMs. >> >>from historic perspective kasparov IS the ideal opponent for a computer. >>if you lose, well it's kasparov. If you win, you beated an ex world champion! >> >>Best regards, >>Vincent > >I am pretty sure that although GK get's a hefty purse win or lose, that he will >be more concerned with winning than anything else... > >I am also sure that DJ will run on more than a dual, maybe an 8-way Xeon? > >"especially taking care he looks good on the photo >instead of in the analysis of the GMs." > >If that was true he would not be the highest rated player in history! sounds >like you are reading something into it, that isn't there... > >Regards >Jonas
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.