AI box: MB & CPU / PCIe lanes required for 4x GPU & 2x NVMe

[keja]

Hello.

 

Im trying to build a box for AI workload (Inference/image recognition) that it is going to run a unraid setup like this:

4x VMs for analysing images. (GPU)

2x VMs for downloading/preparing images for the 4 other VMs (CPU, low load)

 

im just not sure what motherboard and cpu to go for, as im not super much in to hardware anymore, but i do know i would like to add 
4x RTX 2060 SUPER (TU106-410-A1)

2x NVMe SSD 

 

Correct me if im wrong, but i assume that the 4x GPUs can just run with 8x and then the 2 SSDs with 4x (so 40 lanes),

so im looking for any suggestion for a MB/CPU that supports that ?

 

I looked at ASUS WS Z390 PRO, then i realised that i could not fit that with a CPU with enough lanes, as the i9-9900K only have 16lanes, so now im a bit lost.

 

 

Price is not super important, but if i could keep the cost somewhat down that would be great  

 

Thanks in advance.

[Mephi00]

HEDT or zen 2 would be the way to go here because zen 2 runs with pcie 4.0 with double the speed of pcie 3.0...

HEDT chips should have enough pcie lanes run this thing

[keja]

11 minutes ago, Mephi00 said:

HEDT or zen 2 would be the way to go here because zen 2 runs with pcie 4.0 with double the speed of pcie 3.0...

HEDT chips should have enough pcie lanes run this thing

Thanks, just found ASUS WS X299 SAGE - and looks like that could workout with the CPU selection for that card, im not sure about AMD and virtualization, so i will stick with intel and what i know.