4 ms·

Datacenters: memory over fabric (CXL) High End Accelerators: HBM Consumer: maybe on-package DDR like M1 Everywhere: more cache