Bottleneck in E@H computation
24 Jun 2008 9:37:55 UTC
Topic 193733
(moderation:
I was thinking about my various systems, which are relatively close in hardware, but slightly different. Two of the systems have the same CPU, while the third is slightly slower. My Windows XP x64 system has the lowest timings on memory. It's clear that hard drive and graphics performance are not limiting factors here, but what exactly is? Plain-old crunching power? Does memory have an effect? I know that Rosetta@Home does very well with high L2/L3 cache and fast memory, but I'm not sure where that lies here.
Any input is appreciated. :)
Bottleneck in E@H computation
)
Yes, memory does play an important role in E@H performance (especially for those workunits that are "slower" than the average). As a good example, you can look at the "Top hosts" statistics where there is a 16 core AMD system (I think it's "Herr Datenrats" machine). It has a rather unusual runtime characteristic as the relative difference between "fast" and "slow" workunits is much greater than usual. The most likely explanation is that the machine hits memory bandwidth limits (even tho the recent AMD CPUs usually scale a bit better than the Intel ones). So especially for quadcore (and above) systems, E@H likes fast memory & big caches.
Bikeman
Thanks for the info!
)
Thanks for the info!