Based on your current config, I'm guessing budget isn't an issue at all? Are you doing anything besides deep learning? Also, why two separate workstations?

Budget is relatively limited (I wish I could go for a SkylakeX or even few DGX2). It's a 1 WS and 1 server (headless). The WS will be dedicated to deep learning, computer vision and software development. The server will hold a hyperconvergent platform (many VMs) based on Windows server 2019 datacenter; and it will need to stay connected 24/7. The WS should be able to be rebooted or crashed.

Okay. What's your budget? Also, odds are very good that Zen 2 drops during CES (this week), so I'll probably wait until after then to give you a completed answer.

I updated, It's for a 9900k based WS, so no Zen2 for this guy. I'm using loads of code optimized for Skylake and the Intel compiler. Indeed, I wish I could go for a 9940x for instance.

@JMY1000 Budget is fixed since the price(Formula) = price(Hero+NIC 10Gbit). In the first case I got 5Gbit and 32 lines, in the second 10Gbit and 28 lines. That sums up the situation.

Hrm, okay, though I gotta say it seems a little unusual you optimized so heavily for Sky/Kaby/Coffee Lake + Intel compiler that Zen (or even older/newer Intel?) is a no-go. Any chance that could shift? It seems probable that you'll take a massive, massive performance hit because of the limited PCI-e and memory bandwidth on consumer Intel. Also, could you just give a dollar amount please? Prices fluctuate and it's not the easiest to calculate, and it'll make things much easier if I could just have a dollar amount.

Deep learning happens almost fully inside the GPU. Computer vision is more affected by the presence of AVX512, and number of cores. Amd has bandwidth problems at the cache level, and has poor driver/library support. Anyway, here, it's only about the motherboard + NIC.

I'm aware it's almost entirely in the GPU, but you've still got to swap your data sometimes, and given your current plan is to use consumer GPUs, that may be unfortunately somewhat frequently

Obv. you want to minimize how often that happens but it'll happen somewhat

I might add that that price is the same in both case !

What is the $ amount please? Also, what drivers/libraries are you concerned about anyways?

MKL, opencv, Parallel studio XE. 370 (formula) vs 280 + 90 (hero + NIC)

I've had zero issues with OpenCV on AMD. I can't speak for MKL and Parallel Studio, but again, I'm left wondering why you need Intel's specific stuff there so badly.

It's not the point. The point is the MB.

You have some very valid concerns, but I think if you go consumer Intel you're going to have a pretty bad time

I understand. I might really want to consider a skylake X due to the AVX512

Skylake X isn't consumer Intel, it's prosumer/enthusiast

For that, you'll need X299

I already got the server (Intel). I also have the Parallel Studio XE (~4k$.

Yes, I was considering a lot the 9940x + x299 + 128GB instead of a 9900k + 64GB

Beware the sunk cost fallacy, but I see your point.

Okay, so this is just the workstation, yes?

Yes ineed.

indeed.

So, for the same price I prefer a 9900k + 2080ti than a 9940x alone

Okay, I still think we have an issue where you need to be considering the platform as a whole

What would be the 'sunk cost fallacy' ?

That you've already spent money on one solution, so clearly it's best to continue with that solution.