Chinese IT infrastructure solutions provider Inspur and European high-performance computing system integrator Megware have deployed a GPU cluster for a German university.
The system has been installed at the Erlangen National Center for High-Performance Computing ([email protected]) at the Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU).
The 'Alex' supercomputer was part-funded by the German Research Foundation (DFG) as part of INST 90/1171-1. It is named after FAU’s early benefactor Alexander, Margrave of Brandenburg-Ansbach.
The supercomputer features 32 NF5488A5, which are each equipped with eight Nvidia A100 GPUs and two 64-core AMD Epyc 7713 CPUs in a 4U chassis and utilizes an NVSwitch GPU interconnect.
It also includes 38 NF5468A5 Inspur GPU servers, each with eight Nvidia A40 Tensor Core GPUs and two AMD Epyc 7713 CPUs in a 4U chassis and a PCIe 4.0 high-speed interface.
Together, that makes a total of 256 Nvidia A100 Tensor Core GPUs (160x 40GB, 96x 80GB), 304 Nvidia A40 Tensor Core GPUs, 140 AMD Epyc 7713 CPUs, and a total memory capacity of almost 50TB.
Alex is interconnected through a high-speed HDR InfiniBand network.
The supercomputer complements the Regional Computing Centre of Erlangen (RRZE) TinyGPU cluster, a system built out of consumer GPUs of different generations. It also includes eight nodes with a total of 32 Nvidia A100/40GB GPGPUs which may be moved to Alex at a later date.