Servers with liquid cooling slash data centre power costs
Believed to be the industry’s first GPU servers with liquid cooling, the Nvidia HGX H100 eight and four-GPU servers have been announced by Supermicro.
The liquid cooled GPU server contains dual Intel or AMD CPUs and eight or four interconnected Nvidia HGX H100 Tensor Core GPUs. Using liquid cooling reduces the power consumption of data centres by up to 40 per cent, resulting in lower operating costs. In addition, both systems significantly surpass the previous generation of Nvidia HGX GPU equipped systems, providing up to 30 times performance and efficiency in today’s large transformer models with faster GPU-GPU interconnect speed and PCIe 5.0 based networking and storage.
Supermicro’s liquid cooling rack includes a coolant distribution unit that provides up to 80kW of direct-to-chip cooling for today’s highest TDP CPUs and GPUs for a wide range of Supermicro servers. The redundant and hot-swappable power supply and liquid cooling pumps ensure that the servers will be continuously cooled, even with a power supply or pump failure. The leak-proof connectors give customers the added confidence of uninterrupted liquid cooling for all systems.
The liquid cooled Nvidia HGX H100 rack scale servers extend the company’s data server portfolio with liquid cooling technologies entirely from Supermicro designed to reduce the lead time for a complete installation. They also increase performance, and result in lower operating expenses while significantly reducing the power usage effectiveness of data centres, added the company. Savings for a data centre are estimated to be 40 per cent for power when using Supermicro liquid cooling compared to an air-cooled data centre. In addition, up to 86 per cent reduction in direct cooling costs compared to existing data centres may be realised, said the company.
“With the amount of power required to enable today’s rapidly evolving large scale AI models, optimising TCO and the Total Cost to Environment (TCE) is crucial to data centre operators,” observed Charles Liang, president, and CEO of Supermicro.
“These GPU systems are designed from the ground up for rack scale integration with liquid cooling to provide superior performance, efficiency, and ease of deployments, allowing us to meet our customers’ requirements with a short lead time.”
AI-optimised racks with Supermicro product families, including the Intel and AMD server product lines, can be quickly delivered from standard engineering templates or easily customised based on the user’s requirements. Supermicro offers a broad product line of servers and storage systems to tackle complex compute-intensive projects.