Author: Vincent Diepeveen
Date: 12:35:07 08/12/02
Go up one level in this thread
On August 11, 2002 at 21:50:00, Robert Hyatt wrote: Please keep in mind Junior loses less on the dual K7 than crafty does. It is *made* with a dual k7 in mind seemingly. There is of course 1 aspect here which is also important. For all freaks here it is already hard to imagine a dual k7 to be fastest machine for junior. The commercial aspect is however that you get better PR by saying it's a 8 processor box, than that it's a dual. If a guy called FF has a sticker with him at the match kramnik-fritz: "8 cpu inside" And he sticks it to for example at a dual k7 with a very fast WCC 2002 multiprocessing version of Fritz (sold gets the multithreading cpu), then no 8 processor is going to beat it and no one ever will find out. Who's going to be arbiter there anyway? Last rumours i heart from FIDE were that they do not ship an arbiter to the match kramnik-fritz? >On August 11, 2002 at 17:34:36, Chris Carson wrote: > >>On August 11, 2002 at 17:17:09, Jason Jarrells wrote: >> >>> Bunch of idiots here.. Someone asks a simple question, and you give BS >>>statements.. Get a life people. Does anyone know what hardware they are using? >>> If not then don't comment on this question... Geez.. >> >>I do not know and I have also asked this question, perhaps the questions should >>be, what is the fastest multi-proc 4 way and 8 way boxes (Intel and AMD). > > > >Some data from testing myself... > >4-way boxes are very good. With 4-way interleaving, my quad loses about >7% when I test it. IE I run a single instance of crafty, using 1 cpu, >and get a precise time to complete a search. I then run _two_ instances >of Crafty and measure the same time again. It will be about 7% slower >than it was when it ran by itself. If I run three instances of Crafty, >one will run about 14% slower. And four makes this 21%. This is a memory >and bus issue. > >The 8-way boxes still use 4-way interleaving. Which means twice as many >CPUS, twice as many memory accesses, but the same memory/bus speed as the >4-way box. That is _not_ a recipe for blazing speed. > >A typical dual processor will slow down about 30% when the second instance >of Crafty is run, for example, as most duals do not use any interleaving >at all, and the cpus end up competing for memory and starving as a result. > >It is certainly doable to build an 8-way box with 8-way or 16-way (even better) >interleaving. But it drives the cost _up_... > >With the memory issues, Intel has lagged in the quad and 8-way development, >and it might be that the fastest dual is a better machine overall when you >figure in the memory loss and the SMP search overhead loss. For example: > >dual 2ghz would, for Crafty, behave like dual 1.4 ghz processors due to the >30% memory loss, and then with my normal parallel search efficiency of 1.7, >it would look like a 2.4ghz processor. > >quad 700, looks like a quad 553 mhz machine with the 7% per cpu memory loss. >Factor in the 3.1x SMP search efficiency and it looks like a single 1.7ghz >processor. > >Which would you take? > >That kind of math needs to be done for _any_ machine before blindly diving >in and using it in a match this important. No doubt it will be done...
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.