New approaches to improving utilization while reducing guard-banding.
The rollout of generative AI, coupled with more data in general, is requiring data centers to run servers harder and longer. That, in turn, is generating more heat and accelerating aging, and to ensure these systems continue working over their projected lifetimes, chipmakers are building extra margin into chips. That increases the amount of energy required to run and cool them, and it can shorten the lifespan of these systems. Noam Brousard, vice president of solutions engineering at proteanTecs, talks about the need to identify issues involving workloads, aging, temperature, and IR drop in real-time and guard-band selectively on an as-needed basis, rather than trying to protect against a combination of theoretical worst-case scenarios.
Leave a Reply