Does anyone have thoughts on the increased graphics capability of the PENTIUM III as applied to CFD?

Conventional CFD is first of all floating-point number crunching. In this aspect, is Pentium III as good as for instance, Dec Alpha?

Gassan

A good comparison for gauging this is to look at the SPECfp ratings for each machine. You can try www.specbench.org to get SPECfp statistics for just about every machine out there. However, the only reliable way to ultimately gauge the performance you will see is to benchmark the specific code you are using on each machine. The results will vary depending on your algorithm and compiler type/options. We have done extensive benchmarking with our own CFD code here and have found that the relative results usually follow the general trend of the SPECfp ratings. Our ratings show that the Pentium II 450 is about as fast as the DEC Alpha 20264 450. However, the newest DEC ALpha 21264 (or EV6) is quite fast...about twice as fast. Note that the performance on all of these machines will be degraded if your problem is too large to be solved in core memory. We have found that when the code starts disk caching your run times start to double or even quadruple, hence a SPECfp rating is meaningless if your code is inefficient with memory allocation or your machine has very little physical RAM. For these types of problems, IBM workstations are far superior as they have extremely efficient disk IO.

Hope this helps...

Tim,

I'm looking for PC to do 3-D mesh generation and run a flow solver (NPARC/WIND). I'll take your advice and checkout www.specbench.org but do you have more advice on min RAM, hard drive, graphics card, PC manufacturer, etc.