Doesn’t that make sense though as each manipulates a different layer in the memory hierarchy allowing the programmer to control the latency and throughput implications. I see it as a good thing.