NANO 4

• Node specifications: • Processor: Dual Intel Xeon Platinum processors per node • Accelerators: NVIDIA H200 GPUs (8 per node) • Memory capacity: Up to 2 TB per node • Interconnect: InfiniBand NDR 400 Gb/s • System specifications: • Compute node: 220 GPU-based nodes • Additional architecture: Integration with NVIDIA GB200 NVL72 systems • Theoretical peak performance: Over 80 PFLOPS (Rmax measured at 81.55 PFLOPS) • Total storage capacity: Approximately 25 PB • Interconnect: High-speed InfiniBand NDR network (400 Gb/s) • Cooling method: Direct liquid cooling (DLC) • Power consumption: Approximately 2.2 MW • Power usage effectiveness (PUE): Approximately 1.18 • Operating system: • Linux-based (e.g. Red Hat Enterprise Linux) • Workload manager: Slurm ==Installations==

NANO 4 is installed at a high-performance computing facility in Tainan, Taiwan, operated by the National Center for High-performance Computing. The system forms part of a larger AI and cloud computing hub, which was opened on 11 December 2025 and intended to support national research infrastructure and industrial innovation. The system has been used for applications in artificial intelligence, telecommunications, cloud services, and digital content development. It is accessible to academic institutions, government agencies, and industrial users through allocation programs managed by NCHC. ==See also==