OpenFoam Cluster Hardware
 

We're looking to add a number of computers to our CFD cluster and are trying to figure out the best setup to purchase. Ideally we would like to maximize our computation power per dollar without introducing any bottlenecks due to overly subdivided domains and so on.

Currently what we're looking at are dual Intel Xeon E5-2620 chips (6 cores each at 2.0ghz, 2.5ghz with turbo boost, 15mb L3 cache)

The price of the setup that will be running these chips comes out to roughly equal to what we currently buy which run on one Intel Core i7 3930K (6 cores at 3.2ghz 3.8ghz with turbo boost, 12mb L3 cache)

Of course with enough memory to go around

So what I'd like to know is

a. are the xeon chips we chose the best buy for what we're trying to accomplish?

b. because its a "new setup" I'm going to have to present my supervisor with some numbers or evidence that the xeon setup will be superior to the i7 (especially since the price point is give or take equal for the rigs) Is anyone with a similar xeon setup willing to run an openfoam benchmark case that we can duplicate? (2.1.1 or 2.2.0) or know of somewhere with openfoam benchmarks on a similar setup, that would really seal the deal.

The cases we would be running would range between relatively small at 150,000 or so cells on one or two nodes to a maximum of a couple million cells decomposed to 40 - 100 cores. (double the decomposition for the xeon chips so 80 - 200)

It basically comes down to if the 12 cores at a lower clock speed can accomplish what the 6 cores will at a higher clock speed.

Do you have a low latency interconnect such as Infiniband? Or just ethernet?

Unfortunately we're just running on plain old ethernet

Then my guess (!) would be the following: For cases that utilize up to 6 processors, the Core i7 will be faster due to the significantly higher clock rate. For cases with more than 6 processors, the interconnect on the Xeon will outperform ethernet, so you'll start seeing better performance here with increasing CPUs.