FMS 2022 - Power & Latency Analysis of the Memory Tiering Pyramid

The Memory Tiering Pyramid has frequently been used to explain and justify technological shifts in memory architecture over the years. It conveys an idealistic model of graceful progression through memory hierarchy with increasing distance, latency, capacity and persistence. Yet, behind this enticing pyramid is a complex array of architectural nuances that bear little resemblance to the beauty of the Pyramid.
This presentation will break down the memory pyramid into its constituent parts and redraw them with a more accurate representation of today’s computing architectures.
Then the presentation will propose a future Power and Latency friendly Computing Architecture that will faithfully replicate the ideal Memory Tiering Pyramid. The proposed solution will show how both OMI and CXL will play important roles in this Shared Memory Centric, Distributed Computing World.