Abstract

Processor energy density is exceeding the capabilities of conventional air-cooling technology, but two-phase cooling has the potential to manage these resulting heat fluxes at reliable temperatures and higher electrical efficiency. When two-phase cooling is used in tandem with overclocking, data center footprints are reduced as individual chip processing power can be set at limits well beyond the manufacturer’s Thermal Design Power (TDP) or nominal operating condition. This study examines how Liquid Nitrogen (LN2) can be used with Additive Manufacturing (AM) and overclocking to increase the computational performance of a commercially available GPU. The power consumption and frequency relationship were established for both the cryogenically cooled solution and a comparative air-cooled solution. The cryogenic solution saw up to a 17.4% increase in compute efficiency and an 18.1% improvement in compute speed with comparable power efficiency at an equivalent performance level to the air-cooled solution. This study considers the computational performance and efficiency gains that can be acquired through cryogenic cooling on an individual graphics card, which can be replicated on a larger scale in data center applications.

This content is only available via PDF.
You do not currently have access to this content.