Abstract
Datacenters serving artificial intelligence (AI) or machine learning (ML) workloads consume substantial amounts of power and can be well-served by high voltage direct current (HVDC) power lines to enhance rack-level power density. HVDC power systems in datacenters often use a sidecar architecture to provide DC power to the AI/ML payload racks. The impedance of long cables connecting remotely located sidecars to the payload rack gives rise to the potential for resonance and instability. This disclosure describes hybrid energy-storage techniques that combine electrolytic capacitors and supercapacitors within an HVDC system to improve power supply stability and quality. Cable-tray resonance is effectively damped via electrolytic capacitors at rack-tray inputs, and, simultaneously, AI/ML workload power fluctuations are efficiently smoothed via supercapacitors integrated within sidecars. The simultaneous smoothing of power fluctuations and damping of resonant oscillations addresses an important problem in the deployment of HVDC-powered datacenters that cater to AI/ML workloads.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
NA, "Improving Power Smoothing and Resonance Damping in High-Voltage Direct Current Datacenters", Technical Disclosure Commons, (June 26, 2025)
https://www.tdcommons.org/dpubs_series/8284