Advanced Micro Devices Inc (AMD)

[HailMary]

But getting 2 4P 850 systems for EDA work versus the 1 2P IPF system will make the IPF 1/4th the speed for the same cost.

This really depends on the software licensing costs. A lot of EDA tools are licensed per CPU, and the EDA vendors charge enough to overcome any hardware cost differential. It is best to have the fastest single CPU you can to get the most out of the license. In many cases this still means Opteron.

BTW SPECfp does not really correspond to EDA performance as the EDA datasets are much larger.

Now this is an excellent point, and it seems to correlate with what I have personally experienced with various datasets in the >200MB range. The Opteron systems I have used also seem to have unusually high i/o performance, which really helps here. It is obvious that SPECfp is all about cache. Take a look at the various scores of Xeon and P4EE, where the only difference is cache size.