Apart from the SSD and fast memory, aim for a board with the best bus speed possible, and if it’s a Xeon, watch out if it’s got more than one NUMA nodes.
I have just built an i7 6900k, 64gig ddr4 3200mhz, 512gig M2 and 2TB SSD drive. Doing pretty well for simulations. The problem is the memory and I think it is a common problem on x99 boards. To get the memory running at 3200 (XMP) then the board tends to pump more voltage etc. At the moment I am trying to find a balance as it runs far too hot at default levels.
I also have my eye on that 18 core i9 (whenever it is supposed to come out) but in my case, I’m searching for a solution that is great for simulations and also for rendering, which would make the core count more relevant. I think you may be correct that the higher frequency is likely to be better than overall core count strictly for simulations but as Svetlin pointed out, bus speed is a common bottleneck so perhaps it may come down to the motherboard before the cpu.
I’m curious to learn about the limitations of NUMA nodes. Does that refer to the actual number of processors? Does 2 processors mean 2 NUMA nodes? I’ve read Svetlin’s warning that having more than one NUMA can slow things down in the long run but haven’t tested it thoroughly. I’d love to have a benchmark simulation (possibly an approved simulation scene from Svetlin & Chaos Group) that properly stresses the hardware and could be used for comparing configurations by comparing the sim times.