In modern data centers, graphics processing units (GPUs) are the primary load of power infrastructure. All GPUs are co-working with each other, and the power consumption of these GPUs fluctuates rapidly with a synchronized operation, switching between idle and peak load within milliseconds [see reference 2].

These transient load characteristics pose significant challenges for PSUs. The PSUs must respond quickly to the load transient (to keep the output voltages) and maintain current sharing among all the paralleled PSUs during the transient (to avoid overloading the individual power supplies). At the same time, the PSUs must level this power fluctuation on the AC input side with the internal energy storage. In the Mt. Diablo 400 specification, the voltage deviation during load transients must be tightly controlled within ±3% of the output voltage. Additionally, NVIDIA® has presented clear requirements on AC fluctuation during a huge load transient.