Author: Aaron Gordon
Date: 23:13:51 03/19/03
Go up one level in this thread
On March 20, 2003 at 02:00:25, Matt Taylor wrote: >All of these tests were run on an MSI K7N2G-ILSR (nForce 2 chipset) under >Slackware 8.1 using a pc2100 DDR DIMM from Micron. The FSB speed was held >constant at 166 MHz in each test. Just for grins, I threw in a Crafty score. I >ran Crafty thrice for the first test, but after observing that the figure did >not measurably change, I ran it only once on the other tests. I ran lmbench >three times and averages the scores for each trial. > >Mem bus | RAS | RCD | RP | CAS | LM lat (ns) | Crafty score >------------------------------------------------------------ >133 MHz | 6 | 3 | 3 | 2.5 | 177.6 | 884,042 nodes/sec >100 MHz | 5 | 2 | 2 | 2 | 187.4 | 884,042 nodes/sec >166 MHz | 7 | 3 | 3 | 2.5 | 117.8 | 896,671 nodes/sec >200 MHz | 9 | 4 | 4 | 3 | 139.9 | 896,671 nodes/sec > >First, you notice that nForce 2 -really- does not like running the FSB faster >than the memory bus. Second, you notice that Crafty hardly cares about memory >latency or bandwidth. > >BTW, before anyone comments on the abnormally low Crafty scores, I threw >together an unoptimized binary. The same system was pulling >1,150,000 nodes/sec >last week under Windows. > >-Matt Looks like it's pretty clear that running the memory asynchronous to the fsb leads to horrid latency. I noticed the same on my board. 200fsb w/ 100mhz mem was 183ns, 100fsb w/ 100mhz mem was 121ns.
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.