@niconiconi propably...

Q: How much is #MemoryLatency an affecting size vs. the bandwith per DRAM chip?

@fuchsiii speculates that the absurdly low latency due to package-on carrier should yield extreme performance even when accounting the multiplefold density of the few DRAM-ICs on the carrier vs. a DDR5-SODIMM stick...

i've heard from multiple sources that scientific computing today is not bounded by performance time for doing arithmetic but by memory access time and throughput. is there some resource (paper/blog post) that someone here can refer me to so that i can understand it deeply #gpu #gpgpu #cuda #memorylatency