Skip to the content
R. Jenkinsb, A.J. Robinsona,b,c
Abstract
With the continued rise in AI adoption and rollout, GPUs, the computing engines of AI, will continue to increase in power demand and die size. As a result, cooling solutions must also evolve to handle higher thermal loads and changing die architectures. At Nexalus, AI is a core focus, with a dedicated research and engineering team working to anticipate future needs and requirements in this evolving space.
In answer to this AI-driven GPU shift, Nexalus has developed the SoloFlux liquid-cooled cold plate heat exchanger, which boasts unprecedented thermal-hydraulic performance in a compact single-slot form factor.
With the launch of the highly anticipated NVIDIA Blackwell platform – enabling organizations to build and run real-time generative AI on trillion-parameter large language models – this white paper investigates the cooling potential of the Nexalus SoloFlux on this new GPU platform.
Our estimation is based on extrapolation of measured thermal performance data obtained from experiments carried out on the NVIDIA GeForce RTX 4090, giving a realistic indicator of the performance potential of the SoloFlux for the expected power and die size of the NVIDIA Blackwell GPUs.
Copyright © 2024 Nexalus Ltd. All rights reserved.
a Department of Mechanical and Manufacturing Engineering, Parsons Building, Trinity College Dublin, D2, Ireland
b Nexalus Labs, Water St, Loughanalla, Castlepollard, Co Westmeath, Ireland
c CONNECT, Dunlop Oriel House, Trinity College Dublin, Ireland.
