[Echidna]

Hello, i am evaluating the configuration of my next workstation. I am mainly working on FEA (mid-scale assemblies in the order of 4 million elements) but anyway i think you could help me on this.

The 2 possible systems are the following:

1) AMD Threadripper 1950X = 16cores/3.4GHz base clock speed
2) Dual Xeon E5-2680 v4 E.S.2 = 28cores/2.4GHz base clock speed

Let's say that the rest of the hardware (RAM, NVMe or SSD disk) are exactly the same except for the RAM frequency where on the Xeon side cannot exceed 2400MHz i think.

A friend of mine sells both his (ES) Xeons for about half the price of the Threadripper. So, do you think that the higher core count will be of any benefit?

I don't know how indicative is this but the AMD 1950X takes a score of 24967 on Passmark CPU benchmark while the dual Xeon E5-2680 V4 takes the score of 26953.
Even if the 2 configurations are equal in Passmark CPU do you think i should go for the dual-Xeon system because of the higher core count?
And finally given that the Xeons are ES (engineering sample) edition, do you think i could face problems regarding reliability or performance stability? Have you ever used ES Xeons in such demanding environments?

Thanks in advance for your time!

[flotus1]

It is not advisable to use the same memory configuration for both setups.
The dual Xeons shoud be paired with 8x DDR4-2400 reg ECC to make use of the 8 memory channels.
Ryzen TR on the other hand performs best with 4x DDR4-3200 or faster but might be a bit picky with the exact memory model. And using components beyond their official specifications might be undesirable for a FEA workstation depending on the scope of applications and your personal preferences.

should go for the dual-Xeon system because of the higher core count?

Not so much because of the higher core count which is a bonus, but because of the higher memory bandwidth. Scaling with high core counts is usually not that great with FEA.

And finally given that the Xeons are ES (engineering sample) edition, do you think i could face problems regarding reliability or performance stability?

ES CPUs range from barely usable to indistinguishable from the retail CPU. It depends on the stepping i.e. the date when they were produced. Even leaving aside legal issues with these CPUs -they are and remain property of Intel and not for sale- I would not recommend using them for a FEA workstation. Performance will probably identical to a retail CPU, but reliability and stability depend on the unknown type of ES you buy.

The conclusion I would draw is that both setups are not ideal for your purpose. I would either recommend two cheap retail Xeon processors (e.g. E5-2620v4 or 2630v4) or waiting a few more weeks for AMDs "Epyc" server processors.

Edit:

I don't know how indicative is this but the AMD 1950X takes a score of 24967 on Passmark CPU benchmark while the dual Xeon E5-2680 V4 takes the score of 26953.

There are several problems with this benchmark. First of all, it is not very relevant for CFD workloads. But more importantly, it is based on user submissions. As you can see, there is only one sample for the dual-Xeons you looked up and the margin of error is very high. In fact, a single 2680v4 scores 19922 in the multi-core benchmark, not that much less than the reported dual-CPU system.