Warm-water ready
Benchmarking two-phase vs. single-phase direct-to chip cooling on an NVIDIA B200 server
Can two-phase direct-to-chip cooling help AI data centers manage rising GPU power demands while reducing energy consumption?
This whitepaper examines an independent, apples-to-apples comparison of Accelsius’ NeuCool® two-phase cooling system against conventional single-phase liquid cooling on a Dell PowerEdge XE9680L server equipped with eight NVIDIA B200 GPUs.
Across approximately 40K operating points, the benchmark reveals how NeuCool delivers improved thermal performance, lower coolant flow requirements, and greater operating headroom under demanding AI workloads.
Cooler GPU temperatures: Up to 19°C lower temperatures at the downstream GPU cold plate
Reduced thermal resistance: A 48% reduction compared with single-phase cooling
Greater thermal headroom: GPU junction temperatures up to 14°C lower under specified operating conditions
Cooling energy savings: Potential annual savings of 36–56% through warmer facility water and expanded free-cooling opportunities
Tags
Comments
Originally published by Data Center Dynamics. All rights belong to the original publisher.