I run several open-source fully parallel SGX programs on Azure VM with SGX (DC48s_v3, EPC size 256 GB, vCPU = 48 cores, set numTCS=48), but their time consumed are not speeded up much after I use larger thread numbers. E.g., almost all full parallel SGX programs stop at ~9x speedup after 16 threads.
What are the possible reasons that the extra 16 ~ 48-th threads in SGX do not help with the parallel program execution?
