Author: Matthew Hull
Date: 20:46:30 11/25/03
Go up one level in this thread
On November 25, 2003 at 22:10:30, Dann Corbit wrote: >On November 25, 2003 at 21:24:47, Matthew Hull wrote: > >>On November 25, 2003 at 20:57:46, Dann Corbit wrote: >> >>>On November 25, 2003 at 20:03:11, Jeremiah Penery wrote: >>> >>>>On November 25, 2003 at 01:23:27, Dann Corbit wrote: >>>> >>>>>On November 25, 2003 at 00:17:00, Derek Paquette wrote: >>>>> >>>>>>What KN/S is Brutus getting anyone know? >>>>>>Or how far is it searching, or how thorough etc? >>>>> >>>>>From: >>>>>http://www.talkchess.com/forums/1/message.html?330084 >>>>>8 PC cluster, with hardware acceleration. >>>>>2.5 million NPS per card * 8 cards = 20M NPS (less threading loss, which is >>>>>probably considerable). >>>>> >>>>>Might be twice that if there are two cards per machine (it isn't clear to me). >>>> >>>>Listening to Vincent commentate Diep games on ICC, he's been asked about Brutus >>>>a lot. He has said that Brutus does some 600k "software nodes" a second, where >>>>each software node consists of a 2-3 ply hardware search. >>>> >>>>I don't know how accurate this is, but I would assume Vincent has gotten >>>>first-hand knowledge from someone about it and that he wouldn't make this up. >>> >>>I wonder what sort of NPS Vincent is getting with a bazillion processors (doing >>>his little part for global warming -- but then again, we're scheduled for >>>another ice age according to the experts, so he may have just saved >>>civilization). >> >> >>Mridul has been posting the Diep read-outs on ICC. It's running from 2 to 6 >>million NPS. It seems that the longer it thinks, the more processesors get "on >>board" and the numbers start to climb. But at short time controls, the numbers >>stay low. Depths seem to linger around 14 to 16 ply, which for Diep are >>apparently un-precedented. > >That seems low, since it has 512 processors (AFAIR). >6e6/512=12K NPS per processor. >At 100K NPS per processor we would see 50 million NPS, less the SMP loss. I >would think at least 25 M NPS. > >Perhaps they come on in clumps or something, instead of all starting at once. I think Diep must have a very heavy eval. Even on fast duals, it only gets like 150knps at best. He claims about 50x speedup over a PC on the MIPS-based super-comp he's using at 500 processors (460 the last two days). I think he said he averaged 25knps per processor. Matt
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.