The author of this web content explains how to estimate read memory bandwidth using a C++ program. By allocating a large memory region and reading it quickly, one can determine their bandwidth. For multithreaded applications, using multiple threads where each reads a section of the memory region can provide insight into potential bandwidth. The author tested the program on an Amazon server and found it had about 115 GB/s of read memory bandwidth. Plotting the bandwidth against the number of threads used, it was discovered that around 20 threads are needed to maximize memory bandwidth, with most of it achieved using just 15 threads.
https://lemire.me/blog/2024/01/18/how-much-memory-bandwidth-do-large-amazon-instances-offer/