Author: Robert Hyatt
Date: 07:45:00 08/17/05
Go up one level in this thread
On August 17, 2005 at 03:15:20, E. Nielsen wrote: >Hi > >I'm just wondering... > >"...means the search is running about 1/6 its normal performance level..." > >So you are getting 6 times speedup with 8 cpu's. Doesn't that correspond to the >average branching factor? And thus could it be that it's still 6 times speedup >with 16 or 32 or 64 cpu's? > >Eyðun Not at all. The simple formula I have used to approximate speedup has been: speedup = 1 + (CPUS -1) * .7 That is not exact of course, but its a good approximation. I posted some 1, 2 and 4 CPU logs last year, and Martin found that the formula was actually a little pessimistic for that partiular machine and set of test positions (It predicted a speedup of 3.1, but the logs showed an average speedup of 3.3). I have not tested this for large numbers of CPUs, but I have tested it for up to 16 and it works pretty well as a performance predictor. Branching factor has nothing to do with this. the effective branching factor is about 2.5 or so, that is it takes about 2.5x as long to do a depth D+1 search as it takes to do a depth D search, sometimes significantly less...
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.