Author: Vincent Diepeveen
Date: 18:57:07 12/22/03
Go up one level in this thread
On December 21, 2003 at 20:51:19, Bob Durrett wrote: i am writing a huge technical report for the government about diep at the supercomputer. because 512 processors is so much faster than 1, it takes a huge time to calculate the speedups. practical spoken it was worst case about 2 ply deeper in middlegame roughly than a fast pc would have given me (dual opteron). Considering 1 cpu was already 10 times slower practically than a pc processor is, that is pretty good. Some positions it performed real well. Peek was 9.9 mln nps, for diep this is very good. a single processor gives in opening 22k nps and it scaled about 40-60% at the supercomputer sequential (so practically losing from that 22k nps about 11k nps). And that at tournament time control, which nowadays is a lot quicker than some years ago, i do not consider a bad thing. Zugzwang and similar programs like cilkchess ran at similar machines and had despite having a smaller evaluation function, a lot less nps at such machines. Zugzwang at a faster latency T3E with 512 alpha's 500Mhz (that's what memory tells me, all these guys are good in NOT reacting onto emails i ship them) peeked around 2 million nodes a second, i remember it very well. Diep's logfiles just like the huge report i'm writing will be public available. That means that everyone who asks after them will get it. The logfiles you can already get if you want to. The testresults i had will be in the report and the entire outputs are publicly available too. The thing missing is single cpu runs to recalculate testresults which were obtained by overnight runs. One of my conclusions will surprise some researchers (not all of them) and the government will dislike it. So it will be carefully written down. Best regards, Vincent > Hi Vincent. > >Have you had a chance to analyze the GRAZ performance data for your >chess-playing program yet? > >What can you tell us about your technical findings? Did you achieve the goals >you had set for the GRAZ event? > >Someone said they thought you did not have the opportunity to test your software >on the full-up computer hardware prior to the GRAZ event. If so, did that have >an impact? > >Your findings should shed light on the technical aspects of running chess >programs on very large multiprocessor computers [>>8 processors]. > >Bob D.
This page took 0.01 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.