Hi,
What is the proper way to calculate max throughput of memory reads from L2 cache on Maxwell for simple buffer loads, assuming 100% cache hits.
anyone has insight how L2 works or can point to a whitepaper?
I propose you use this micro-benchmark tool suite:
Its goal is to expose bandwidth of fast on-chip memories. Try “cachebench” tool which aims to assess L1, L2 & texture cache bandwidth.