[Stabum]

Hello
I would like to build a small CFD machine to be used for CFX and FLUENT simulations.
I firstly thought that, in order to achive the highest computational power, I should have maximized the cores number.
With this statement in my mind I've conceived this ideal machine:

Motherboard: Supermicro MBD-X10DAL-I-O ATX Dual CPU
Processors: 2x Intel Xeon E5-2620 V4 2.1 GHz 8 core (obviously @ stock frequencies)
Memory: 2x(4x8GB) DDR4 2133Mhz
CPU Cooler: Ordinary Air CPU Coolers

Then I remembered that, right in this website, I've been said that the memory operating frequency, even more than the cpu frequency, is the key factor (CFD is probably the unique task where the RAM frequency is so important). Xenon ECC memory cannot be oveclocked (as far as I know), that's why I turned to the Intel's i7 processors, and I've conceived this other machine:

Motherboard: Asus X99 Deluxe II
Processor: Intel i7 6900K overclocked @ 4200MHz
Memory: DDR4 4x16GB 3200MHz C14
CPU Cooler: Custom Liquid Cooling

Maybe it's a stupid comparison, I know, but I'm so curious... In your opinion which one of them will be the fastest? And WHY? Please don't take into account factors like noise, energy consumption...

Thanks SOOOO much in advance!

[flotus1]

So you have enough parallel licenses for 16 cores?
Then choose option 1. More cache, more memory bandwidth. Try getting dual-ranked DDR4-2400 reg ECC, it costs about the same as DDR4-2133 reg ECC and if you ever buy better CPUs you can keep the memory.
The Asus Z10PE-D16 WS motherboard might be the right choice for a hardware enthusiast. It has some pretty nice performance-related features you will not find on other dual-socket motherboards.

[Stabum]

No, no... Actually I don't have such licence... It was just a mere comparison: since you can buy a 8core cpu paying human prices I was curious to know whether a 16 core machine with ECC memory would be faster than the almost fastest single core machine. Thanks so so much!

[pawl]

Hi,
I actually have slightly below your system
I have:
- i7 6900k (not overclocked, just put it to turbo from bios, than get 3.8-3.9 GHz)
- x99 sabertooth bios 34xx
- ddr 4x8GB 2400MHz
- tower dual fan

Previously i put xeon e5 2620v4, and I'm sure this 6900k give better job.

but if you want to spend more, I recommend you to buy gpu tesla series, since you can distribute some computation like described in here https://www.youtube.com/watch?v=9YH9p2KbRls

I have gtx 1080, and when I tried to distribute only 1 computation to gpgpu in Fluent (VOF, turbulence), it made slower, and sometimes give wrong result.

[Stabum]

The Xeon e5 2620v4 was in single configuration or you had a double socket with two CPUs?

[pawl]

Originally Posted by Stabum

The Xeon e5 2620v4 was in single configuration or you had a double socket with two CPUs?

single actually, on my sabertooth x99