Author: Vincent Diepeveen
Date: 17:15:44 09/19/03
Go up one level in this thread
On September 19, 2003 at 16:56:21, Derek Paquette wrote: >On September 19, 2003 at 11:14:29, Vincent Diepeveen wrote: > >>Hello, >> >>I'm looking for positions to search very deep with a cpu or 500 which are >>interesting to search deep. Seemingly a whole machine must get reserved for DIEP >>world champs 2003. The NWO/NCF perhaps needs to pay for the cpu hours also >>during the night says SARA (there is 7 organisations involved and i have written >>at least 1 page for each processor to get it so i won't even start quoting all >>the organisations involved). >> >>Most jobs run like 2 weeks or so at say for example 32 processors, so all those >>jobs cannot get started trivially in the night. >> >>Anyway more negotiations will be there, because the world champs like this will >>cost 80000 cpu hours. >> >>This is all internal talks. In case any of such scenario happens or even between >>rounds when diep finishes sooner its games, there sometimes is a few hours left >>to run diep at 500 cpu's at for example 1 position each run. >> >>that's 250Ghz with around 200GB hashtables in total. >> >>So this allows some massive deep calculations. I'm looking for cool positions to >>try. >> >>Please don't mention the openings position, i'm already using that now and it >>isn't very interesting at all to search that one deep. >> >>feel free to email to me about this: diep@xs4all.nl >> >>Best regards, >>Vincent > >How much of a true depth are you getting compared to say Fritz or >Shredder...could you have us an idea of how many more positions your DIEP will >be looking at with this setup compared to programs running on desktop computers? First of all Fritz doesn't have a clue about good/bad bishops to start with, it's very well tuned at the few positions it plays though. Nimzo indian for example where black doesn't have bishop pair, so it not knowing about bishop pair without losing too many games is explained easily. But i hope you realize what 500 processors with 512GB ram from which i'll take about nearly half as hashtable (whatever type of hashtable) can do at real slow levels. So like a few hours a position. The effect at tournament level trivially is not so devastating like after a few hours. there is simply no scaling problems with such big hashtables. branching factor only goes down and down. Best regards, Vincent
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.