It’s simple and quite forward question,
since we gonna have to make a lot of calls to the memory, and not using a lot o bandwidth anyway (as far as i can see), does LATENCY come into play here? Is DDR3 better than GDDR6 and HBM2?
Latest GPU miners use primarily bandwidth and memory latency is almost completely hidden. If you want to build lean-mining FPGA then DDR3 with as many channels as possible might be a good idea as you need to access single bits.