Abstract
Data centers are one of the fastest-growing loads on the grid, and with the advent of machine learning techniques such as large language models, there is an ever-increasing need for increased computing power. Although modern CPUs and GPUs continue to achieve energy efficiency improvements, there seems to be an insatiable appetite for more and more data processing, as AI tools become common in our everyday life. Single data centers can consume tens of MW of electricity and require sophisticated power distribution, with single racks consuming tens of kW of power, and GPUs surpassing the 1 kW point for a single chip. The power conversion from the high-voltage AC, to the eventual CPU/GPU voltage of less than 1 V, requires continued innovation and new approaches to achieve compact and highly efficient solutions.
| Original language | English (US) |
|---|---|
| Title of host publication | Wide Bandgap Power Electronics |
| Subtitle of host publication | Emerging Converter Technologies and Applications |
| Publisher | Springer Science+Business Media |
| Pages | 493-546 |
| Number of pages | 54 |
| ISBN (Electronic) | 9783031786310 |
| ISBN (Print) | 9783031786303 |
| DOIs | |
| State | Published - Jan 1 2025 |
All Science Journal Classification (ASJC) codes
- General Energy
- General Engineering
- General Physics and Astronomy
Fingerprint
Dive into the research topics of 'Data Center Power Delivery: Capacitor-Based Power Converters'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver