Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Rybka Benchmarks [L2 cache testing]

Author: Aaron Gordon

Date: 15:21:48 12/07/05

Go up one level in this thread


It appears Rybka is no different than other engines when comparing L2 cache
size. I'd like to test on larger cache chips though sometime. Anyway, here is
what I have so far.

Both cpus have the exact same core, the only difference is cache size. Both cpus
were tested in the same computer so everything but the L2 cache size was the
same.

Athlon (Thunderbird) 800MHz, 128k L1 cache, 256k L2 cache:
3988843 nodes in 106.394 seconds = 37494.76 NPS

Duron (Spitfire) 800MHz, 128k L1 cache, 64k L2 cache:
3988843 nodes in 109.579 seconds = 36401.52 NPS

So, going from 64k L2 (192k total on-chip cache) to 256k L2 (384k total on-chip
cache) yielded a 3% increase in performance. Perhaps cache speed makes more of a
difference instead? The only chips that are relatively identical except for the
cache speed (that I can think of) are the Mendocino Celeron 533 and the
Coppermine Celeron 533. I do not have either of these chips though...

I'll try doubling memory speed on my Athlon 64 (single channel at 3.5gb/s vs
dual channel at 7gb/s) and see how that changes Rybka performance. If it doesn't
do anything (or much), then I'll have to guess that cache speed is the factor
behind the Celeron-P4 core being significantly slower than the standard P4 at
the same clock speed.

Anyone have any ideas/suggestions?



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.