Author: Aaron Gordon
Date: 09:41:25 11/01/05
Go up one level in this thread
>Thanks, Günther. A disappointment as P4 is concerned... > >Regards, >Gábor As far as HT goes, please take Hyatts advice there. I don't know the exact numbers, but here is a decent example. Lets say you've got a dual CPU system where 1 cpu takes 60 seconds to hit 16 ply. As soon as you switch to 2 threads/cpus, you get a 20% performance decrease. So, lets figure two cpus that can do 16 ply in 60 seconds a piece, minus 20%. That puts us at ~36 seconds, a nice improvement over 1 cpu. Now, with HT, it isn't a real second CPU. So, lets say we take that same performance decrease and see what it does with HT enabled. 60 seconds per cpu again, 1 cpu. HT will probably benefit you about 10%, but, going into SMP mode again drops you 20%. So, you actually lose performance. Now your 60 second time turns into about 64.8 seconds (if I did my math properly, heh). Here are some benchmarks I ran to compare a few systems. This is with Crafty 19.15 compiled by Bryan Hofmann (http://cito.atspace.com/) run with the "bench" command and 48mb hash. I would have enabled HT for the P4 test, however, my P4 does not have HT. 3054kn/s - Athlon 64 2.85GHz (San Diego) 64 bit 2149kn/s - Athlon 64 2.85GHz (San Diego) 32 bit 1934kn/s - Athlon 64 1.80GHz (San Diego) 64 bit 1612kn/s - Athlon XP 2.50GHz (Barton) 32 bit 1350kn/s - Athlon 64 1.80GHz (San Diego) 32 bit 1209kn/s - Pentium 4 3.0GHz (Northwood) 32 bit 1095kn/s - Athlon XP 2500+ 1.83GHz (Barton) 32 bit
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.