Author: Robert Hyatt
Date: 14:56:25 06/14/03
Go up one level in this thread
On June 14, 2003 at 17:44:33, Brian Richardson wrote: >On June 14, 2003 at 12:36:43, Sean Mintz wrote: > >>Here are my results (which is why I wonder about the quality of the >>executables)... >> >>dual 1.34 ghz athlon mp's: >> >>--cut-- >>EPD Kit revision date: 1996.04.21 >>unable to open book file [./book.bin]. >>book is disabled >>unable to open book file [./books.bin]. >> >>Crafty v19.3 (1 cpus) >> >>White(1): mt=2 >>max threads set to 2 >>White(1): log on >>White(1): hash 384m >>hash table memory = 384M bytes. >>White(1): hashp 24m >>pawn hash table memory = 24M bytes. >>White(1): bench >>Running benchmark. . . >>...... >>Total nodes: 59491829 >>Raw nodes per second: 1451020 >>Total elapsed time: 41 >>SMP time-to-ply measurement: 15.609756 >>--cut-- >> >>Exact same test as you and my 1.34ghz athlons beat your 1.6 ghz athlons. >> >>You can get the executable I used from >>ftp://newageoc.com/pub/crafty/c193smp-k7sse.zip > >I think the elapsed time is more important than nps. >The Crafty from Bob's site ran in 39 seconds. >I also used the executiable you cited and ran as follows: >Total nodes: 59000520 >Raw nodes per second: 1594608 >Total elapsed time: 37 >SMP time-to-ply measurement: 17.297297 It depends on _what_ you want to measure. total elapsed time is affected by random chance in a SMP search environment. It can vary significantly, which means comparing this between machines or 1 cpu to 2 cpus is not a good comparison if all you want to know is "how much faster is the two cpu hardware?" NPS is the way to compare _hardware_. The faster the overall NPS, the better, when using the _same_ program on different hardware platforms. If you want to measure search time, that is certainly a useful comparison, but then you can't use that to predict how _different_ applications might do on a dual compared to a single cpu machine, while the raw NPS number is very useful in comparing that. But of course, then you have to use the _same_ compiler and everything else, or there is another degree of freedom in the resulting data...
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.