Hewlett Packard Enterprise this week said it had landed an order to build a new supercomputer for the National Supercomputing Centre (NSCC) Singapore. The new system is powered by AMD's Epyc 'Milan' processors as well as Nvidia's A100 compute GPUs; it is eight times more powerful than its predecessor.
NSCC's supercomputer uses HPE's liquid-cooled Cray EX architecture and will be one of the world's first systems based on AMD's 3rd Generation Epyc 'Milan' processors based on the Zen 3 microarchitecture (around 100,000 cores in total) as well as 352 of Nvidia's A100 compute GPUs.
The system will be comprised of about 900 nodes and is projected to provide an aggregate of up to 10 PFLOPS of raw FP64 compute power. The NSCC and HPE do not break down CPU and GPU performance in the system, though 352 Nvidia A100 GPUs can offer roughly 3.4 FP64 PFLOPS as well as a whopping 110 FP16 PFLOPS. Meanwhile, the system will be accompanied by a 10PB Cray ClusterStor E1000 storage system with over 300GB/s of read/write performance speeds.
The combination of AMD CPUs and Nvidia GPUs makes it possible to use the new supercomputer both for artificial intelligence (AI) and machine learning (ML) as well as traditional high-performance computing (HPC).
Initially, the NSCC will use its new HPC machine for biomedicine, genomics, diseases, engineering, and high-resolution weather modeling, but over time it can be used for a wide range of applications that require AI and/or HPC.
The NSCC and HPE expect the new supercomputer to be operations sometimes in early 2022. The system costs SGD $40 million (around $30.16 million).