In this talk we will explore the power delivery designs that make it possible to maintain voltage stability on the latest High-Performance-Compute (HPC) chips such as AMD SP5, Nvidia H100 and the upcoming Blackwell B200. 

The latest HPC chips can go from demanding 100A to 1000A near instantaneously which proposes a monumental challenge for Voltage-Regulation-Modules (VRMs) that have to respond in real time to this load change and keep the delivered voltage within tight tolerances.

We will discuss why this level of power is required, the technical limitations that make it challenging and the techniques used to overcome this barrier.