BRK436 Breakout Wed 18 Nov 09:00 PST
Scaling Azure infrastructure: From intelligence into enterprise impact
Saurabh Dighe
Summary
From the recording / Microsoft AI summary
Power, cooling, memory and capital were named as the constraints on scaling Azure infrastructure. New high-memory M-series VMs, in preview, reach 8 TB on two sockets, with 26% better performance per vCPU and 47% less processor power.
Session description
Microsoft's catalogue description
Scaling Azure infrastructure for AI takes more than adding compute, memory and capacity. It requires maximizing “useful yield”: optimizing how efficiently custom silicon, memory, networking, power, cooling and datacenter design are co-engineered for performance per watt, higher throughput, and lower latency. Join us for an inside look at how Microsoft designs and operates end-to-end to turn AI infrastructure into predictable, efficient, and affordable intelligence for the next frontier.