I assume that you have been looking at the Administration forum for information on the best processors and are asking here to get specific information about the demands that simulation makes on the processor. Apparently, no one here has any good answers. I don't either, but I hope my reply will jolt someone who knows into giving you some better information.
My guess is that the two most important parameters are clock speed and number of cores. I'm not sure how well Simulation handles multiple cores. I seem to recall reading here that it depends upon which solvers you are running. With most simulation software, the effect of adding more cores drops off as the total number of cores increases. Going from 1 to 2 may nearly double your speed. Going from 2 to 4 may give you a smaller increase, Going from 4 to 8 will probably give you an even smaller increase.
It would seem like cache might be at least somewhat important, since FEA does a lot of the same steps over and over again, but I have no idea if that really matters in practice. Perhaps the chunks of code and data that it uses fit well in a reasonable amount of cache and it makes a big difference. Perhaps the chunks are too large and cache does not matter at all.
Since FEA uses lots of memory, I would think that memory bandwidth would be important, but that is only my guess. I have no benchmark data to back that up.
I think ECC is a religious issue. Some people feel very strongly about the need for trusted memory and others seem to feel just as strongly that it is a waste of money. I suspect that that the same goes for Xeon vs "consumer" processors. My own feeling is that Intel and the PC makers are very good at extracting more money from "professionals" than from "consumers" for fairly small benefits, but I've got no proof.
If you tell us what motherboard you want to put the CPU in, someone here or in the Administration forum can probably tell you what would be best or tell you that you should really be considering a whole new computer.
Thank you for your input. Our "IT guy" already purchased the computers a couple days ago, so my question is moot now. But I'm still interested from an academic standpoint.
best performance improvement for simulation is a well simplified model and good setup.
next is clock speed.
next is making sure your problem stays in ram.
next is SSD.
then you can start throwing cores and better cache at it.