Hello, I was wondering if NFS was one of those projects like PrimeGrid where cache size is important and we should try to not exceed the cache memory our CPU has by limiting the amount of tasks done at one. Also, is using hyper-threads (use 99% or 100% of the CPU cores in Boinc Manager) recommended or should we stick to use only physical cores (use 50% of the CPU cores in BM) to optimize throughput?

I'm not an NFS@Home expert, but my sense is that it is quite good when during the sieve phase you end up getting hits in L1 (which ggnfs - the siever in NFS@Home - is designed to do), but there are other phases (TD, other factoring of larger numbers), so normally when you use hyperthreads you should see a speedup despite each thread being designed to use the entirety of the L1 cache given to a core because it's only during the sieving phase that it's important.

So feel free to use your hyperthreads! Let us know if it doesn't seem faster, but it should in aggregate.

Sieving loves HT. Here BOINC only sieves. All on. Rule of thumb is that you get 30% more output.