The default answer is the new Threadripper - which doesn't have a confirmed release date yet, but should be out within the next 3-6 months. However, if you really care about optimizing for that workload, the exact answer depends on the codebase(s) you're dealing with because the time spent linking varies significantly.

For instance, if you're working on a large C++ codebase with incremental compilation, rarely do full recompiles and heavily rely on link time and profile guided optimizations, I can see a case where a heavily tuned Cascade Lake X makes a lot of sense (overclock the mesh, tune the memory latencies and get enough cooling to sustain the single core boost as high as you can). It's actually one of the few usecases (AVX-512 being another) where I still see a niche for Intel CPUs purely based on technical merit... saying that still feels weird, but that's where we are now.

I'd like to point out that Threadripper using Ryzen 3 has not been confirmed or denied by AMD. The only dribble of information we have is Threadripper was officially removed from the roadmaps and anonymous sources saying it will still be released. The primary source for our Threadripper dreams says as much: https://wccftech.com/exclusive-amd-is-working-on-a-monster-6...

It'd be stupid for AMD to not release another generation of Threadripper with up to 64 cores, but we don't know anything for sure just yet except what has been detailed about the consumer Ryzen and the server EPYC.

AMD said Threadripper 3000 is coming; I speculate there will be 24/32/48/64-core SKUs.

That'd be my guess as well... I wouldn't that be surprised to see a 16 core entry too (get onto TR platform and upgrade later option). Though the $750 for a 3950X is already pushing my budget, I might consider a 24-32 core for under $1200 if that's where it lands. System budget around $3500.

There will apparently be a 64 core threadripper.