Trying to pick system
 

So I'm looking to run optimization for several different sized fans. I've already done a few individual runs using Fluent RKE at about 8 millions cells (24GB RAM using double precision). Using a single core (5 year old Xeon) it takes about 2-3 days for a single run. I would like to build a machine which can get this down to about 8 hours.
I've got a budget of up to $5000, but that might be overkill and would prefer to stay under.

I've put together a system, feedback would be appreciated. Because I'll be using multi-core I'll be ditching Fluent for something like OpenFOAM so I don't have to worry about the added cost of Fluent.

CPU: Intel Xeon E5-2650 v3 Haswell 2.3GHz Link
MoBo: SUPERMICRO MBD-X10DRI Dual 2011 R3 Link
RAM: Thinking of something like this

Also a few questions:
1) The latest generation of RAM (DDR4) has a high Cas latency (15 ms) and that makes me think I should just go with a slightly older 2011 v2 board with DDR3 ram. Is the lower latency better or is the newer DDR4 worth it?

2) I initially want to start with 32GB of RAM, should I go for 2 sticks of 16GB or 8 sticks of 4GB? Is there a method for maximizing the QPI's transfer rate?

Thanks.

The CAS latency has the unit of cycles, not seconds.
So DDR4 2133 MHz with a CL of 15 has about the same latency as DDR3 1866 MHz with a CL of 13. But dont worry too much about latencies, for CFD applications you will want to maximize memory bandwith instead.
Which brings us to your second question: The Xeon V3 CPUs all have four memory channels, so using only 2 DIMMs per processor will effectively halve the performance you get. Use at least four DIMMs per processor, so at least 8 in total if you want to use 2 processors.

The processor you picked seems like a good choice, it is the cheapest that officially supports 2133 MHz DDR4