By nature HFT is not highly multithreaded and avoids handoff.
So having more cores is not that different than having more discrete CPUs -- no benefits for cache locality/faster coherency.
In that regard if latency is an issue overclocked 8700K (or 8086K) should be sufficient.
Binned & overclocked i9-9980XE’s are already available though. I’m honestly at a bit of a loss to understand Intel’s latest offering. But hey, it made us stop laughing about 10nm for a few minutes.
It is, but at some point you can't do much more about the network latency; shaving off a couple dozen us, might the difference of getting the price/quote or missing it out. Especially, when you play versus similarly equipped firms.
Depends on how low the latency already is. If you're already <0.5ms away network wise then perhaps shearing off 0.1ms from a 1ms operation could make a significant difference.
I'm guessing the chance of an eventual calculation error is not worth the potential speed gains (though you could run those Pi calculators to check for CPU stability)
HFT is not so much about CPU speed but algorithmic optimization.