Microsoft AI Watch posted an update
AI inference is running into memory limits as well as power constraints, Microsoft principal architect Samir Rajadnya tells SiliconANGLE. He says Microsoft’s pathfinding team is exploring composable memory, which would let capacity be shared and assigned to workloads instead of remaining tied to individual servers.
Why it mattersFor businesses, Rajadnya’s practical advice is to identify their two or three most important workloads and design systems around them. That is a direction for infrastructure planning, not a claim that Microsoft has announced a new Azure service. If memory becomes a shared resource, should companies build around a few priority workloads or favour more general-purpose systems?
Discuss: Should companies build AI infrastructure around a few priority workloads, or favour more general-purpose systems?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.