Or using Blade Servers.

THere is another explanation for several hundreds of CPU cores.
On FAH forums it's common knowledge that multi-core CPUs sometimes run into issues with distributing the load.
One core feeds the remaining cores, much like with Nvidia GPUs; but certain core configurations don't work very well with FAH; and there's an upper limit on the amounts of threads per CPU that can be entered.
 
One can however, add 1 CPU slot, using x-amount of threads (say 16), and add an additional CPU slot using an additional (say 16 threads), and an additional, and....;  even if one only has 1 actual CPU core.
If one has a server with say, an AMD Epyc CPUs (64 cores, 128 threads), FAH could register 4 slots per EPYC CPU, at 16 threads per WU (or CPU slot);
One would expand this linearly; eg: have 10 of these CPUs, you could be running near to 40 CPU slots.
 
Or, add 1 CPU slot per thread, and it could show up as 640 individual CPU slots, even if in actuality the CPU has only 10 real CPU cores.
And about 25-33% More, if you enable the hyperthreading.
 
Still, 10 Epyc cores at a rough estimate of ~1M PPD per CPU*, would only amount to 10M PPD; not the 100M PPD reported in eoc...
*(this, derived from it's 16 double precision flops per cycle, per core specs rating; which means it probably will do 32 * 64 * 3000 = 6.144 Tflops, which is very close to an RTX 2060's 6.451 Tflops and 1.04M PPD's rating)
 
I don't really see the benefit of this kind of configuration, over, eg: using 16 or 32 threads per CPU slot.
Running one WU per thread is giving more overhead, than when combining more threads per WU...
 
It's just a theory though... I haven't actually tried running these kinds of CPU workloads, though I could try with an older 10 core, 20 thread Xeon processor I own, to test..