Supermicro announces the upcoming availability of a new, modular GPU server. The server supports multiple compositions of GPUs and CPUs. Freedom of choice is key: the introduction aims at a flexible, optimal fit for various high-performance computing and AI applications.
The aptly named ‘Universal GPU Server’ has a modular design. Supermicro is launching the server to provide clusters of servers with the CPU and GPU power for HPC and AI applications.
At the detail level, said applications have varying demands. The optimal balance of CPUs and GPUs in a server cluster varies accordingly. Hence, Supermicro puts freedom of choice at the heart of its GPU server. Both Intel Xeon Scalable and AMD EPYC processors are supported; AMD Instinct MI200 and NVIDIA A100 GPUs idem.
The introduction is part of the bigger picture. Supermicro strives to deliver the hardware needed for HPC and AI applications to organizations as a complete package. The perfect profile of ‘hardware needed’ is different for each application, necessitating modular designs.
The new Universal GPU Server is geared to connect to Supermicro’s existing HPC system designs (clusters, ed.). Although Supermicro has traditionally cooled its hardware with air, the organization argues that fluid cooling is important for servers with the latest CPUs and GPUs. Hence CEO Charles Liang emphasizes Supermicro’s attention to the technology. “We can now supply large numbers of liquid-cooled clusters with the latest computing technologies.”
The moment of the Universal GPU Server’s release in Supermicro’s systems is unknown at this time.