Supermicro Unveils Liquid Cooled NVIDIA HGX H100 Servers

Supermicro liquid cooling

Supermicro, a global server vendor and provider of total IT solutions for cloud, AI/ML, storage, and 5G/Edge, has introduced NVIDIA HGX H100 servers with liquid cooling technology. These advanced rack scale solutions would offer accelerated deployment, increased performance, and significant reductions in operating expenses, resulting in “up to 40%” lower data center power costs.

The integration of Supermicro’s liquid cooling technologies streamlines the installation process, enabling faster deployment of complete liquid cooled systems. In addition to enhancing performance, these solutions would substantially reduce the power usage effectiveness (PUE) of data centers. Compared to traditional air-cooled data centers, according to the vendor itself, Supermicro’s liquid cooling solutions can deliver an estimated 40% savings in power consumption and up to an 86% reduction in direct cooling costs.

“Our innovative GPU servers, powered by liquid cooling technology, significantly reduce the power demands of data centers,” said Charles Liang, President and CEO of Supermicro. “Given the high power requirements of today’s large-scale AI models, optimizing the total cost of ownership (TCO) and the total cost to the environment (TCE) is critical for data center operators. We possess proven expertise in designing and building high-performance server racks. These GPU systems are purpose-built for rack scale integration with liquid cooling, delivering superior performance, efficiency, and ease of deployment, thereby meeting our customers’ requirements within a short lead time.”

Coolant Distribution Unit

Supermicro’s GPU servers, including the latest Intel and AMD server product lines, are optimized for artificial intelligence (AI) workloads and can be quickly delivered as AI-optimized racks. These racks can be customized based on the unique requirements of users, offering a comprehensive and high-performing solution for compute-intensive projects. By offering fully integrated solutions, Supermicro would simplify the deployment process for customers, allowing them to become productive faster.

The flagship liquid cooled GPU server from Supermicro features dual Intel or AMD CPUs and either eight or four interconnected NVIDIA HGX H100 Tensor Core GPUs. By utilizing liquid cooling, data centers and server operators can achieve a significant reduction in power consumption, resulting in lower operational costs. Furthermore, according to Supermicro, these server systems would outperform previous generation NVIDIA HGX GPU-equipped systems, delivering up to 30 times more performance and efficiency for today’s large transformer models, thanks to faster GPU-GPU interconnect speeds and PCIe 5.0-based networking and storage.

The liquid cooling rack level solution from Supermicro consists of a Coolant Distribution Unit (CDU) that provides up to 80kW of direct-to-chip (D2C) cooling for today’s highest TDP CPUs and GPUs for a broad variety of Supermicro servers. The redundant and hot-swappable power supply and liquid cooling pumps guarantee that the servers will continue to be cooled even in the event of a power supply or pump failure. The leak-proof connectors would provide users with the confidence that all systems will continue to receive uninterrupted liquid cooling.