Author: Robert Hyatt
Date: 12:04:59 07/10/98
Go up one level in this thread
On July 10, 1998 at 12:33:42, Tom Kerrigan wrote: >On July 09, 1998 at 21:32:53, Robert Hyatt wrote: > >>My first benchmark on a PII/300 showed 1.41 times faster than a P6/200. That >>missing .09 (20%) is caused by the 1/2 speed cache on the PII. If you take this > >The PII/300 has the same exact front side bus speed as the P6/200. A chess >program is CPU *and* memory bound, so you can't expect performace to scale >exactly with clock speed. > sure you can... if you know what your program is doing WRT memory accesses. 256K L2 works well with Crafty, holding the engine core and often used memory stuff. 512 holds nearly everything... with the major exception of hash tables, which no cache will help much with. >>up to the PII/400, you begin to see a significant performance loss when compared >>to a P6/200, factoring in the 2x faster clock not giving anywhere near 2x the >>cpu performance. Xeon will. The 450 should clock in just as you'd expect with > >Your measurements simply prove that Crafty is very dependent on memory. >Remember, the PII L2 cache *always* runs at half the core clock speed. >If Crafty fit in L1/L2 cache, you *would* see a ~2x speedup. The PII doesn't always run at 1/2. That's what the PII/Xeon is all about, basically back to 1:1 between L2 and cpu... and benchmark results that I have done have been perfectly linear between the PII/Xeon and the P6/200 that I use daily... finally... The first PII's (before xeon) running 2:1 cpu to cache clocks were pretty close (1.41 compared to optimal 1.5 on the 300 to 200 comparison) but the xeon looks good from my perspective. And note that these are *real* results from data obtained from 450mhz xeon results, and comparing to my P6/200 with everything identical. > >Cheers, >Tom
This page took 0.03 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.