Author: Robert Hyatt
Date: 09:29:11 02/10/03
Go up one level in this thread
On February 09, 2003 at 23:36:05, Tom Kerrigan wrote: >Sucks to have those tables, I guess. > >How does Crafty scale given different cache sizes? > >-Tom I don't know, as all I have been able to use is 256K, 512K and 1024K. I did run on some 2048K xeons at one point, and they seemed to scale a bit better, but not a lot, and certainly not worth the nearly $5,000 per chip (at the time). In looking back at old data, the 1M 700mhz processors I had did scale better than the 512K 400 and 512K 550mhz xeons I ran on. IE the 700mhz X 1mb were well over 1.5X faster, and closer to 2x faster. Unfortunately there are some "family" issues in this as well (PII vs PIII) so it isn't so easy to say that the 1M L2 cache was the only thing that helped. But for comparison, a PII/400 (512KL2) runs the benchmark at 630K nodes per second. a PIII/550 (512KL2) runs the benchmark at 988K nodes per second. a PIII/700 (1024KL2) runs the benchmark at 1335K nodes per second. 550/400=1.38X 988.630=1.57X 700/550=1.27X 1335/988=1.35X So I guess, for those last two machines, that the double-size L2 has little effect. The first comparison is a PII vs a PIII so part of the speed difference came from the coppermine upgrade. I don't have access to the 2M L2 machine as I simply evaluated it for someone here on campus and we concluded "not worth the cost". It was better, but not _that_ much better to justify almost $16,000 more (the 1M 700s were $1300 at the time). The 2Ms were nearly $5,000. And they were not in plentiful supply either.
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.