Aspire 2A+

Powering AI-Driven Science and Innovation

ASPIRE 2A+ consists of NVIDIA DGX SuperPOD with NVIDIA DGX H100 systems, providing advanced accelerated computing capable of supporting research in areas such as AI, large language models and machine learning as well as advanced materials research, biomedical and health sciences under the AI for Science effort. With an estimated 20 PFLOPS of raw compute power at FP64 Tensor Core precision, it will significantly enhance Singapore’s ability to lead research in emerging technologies.

Other Platforms
Specifications
  • A DGX SuperPOD with 40 DGX H100 systems, containing a total of 320 NVIDIA H100 GPUs
     
  • 400 Gb/s NVIDIA InfiniBand (Networking) 
  • 2TB (System Memory) per system 
  • 27.5 PB (Home Storage)
  • 2.5 PB (Scratch Storage)
AMD-based Cray EX supercomputer
  • 105,984 Cores (CPU - AMD EPYCTM 7713)

  • 1,024 Cores (AMD EPYCTM 75F3)

  • 352 NVIDIA A100 Tensor Core GPUs

  • 476 TB (Total System Memory)

  • 25 PBytes Storage (Spinning + Nearline)

  • 10 PBytes (Scratch Disk)
AI System
  • 18 AI GPU nodes

 

  • 12 nodes with 4x Nvidia A100 40GB and 12 TB nvme local storage

 

  • 6 nodes with 8x Nvidia A100 40GB and 14TB nvme local storage
High Frequency
  • 16 DL385 High Frequency Nodes

 

  • Dual-CPU AMD 75F3 (32 cores/CPU + 32 cores/CPU = 64 cores in a node)

 

  • 100G High speed network

 

  • 512GB DDR4 ECC RAM (User accessible RAM = 500 GB)

 

  • Red Hat Enterprise Linux-8
High Speed Network Interconnect (HPE Slingshot)
  • All nodes are connected with HPE Slingshot Interconnect (Dragonfly Topology)

 

  • HPE Slingshot provides a modern, high-performance interconnect for HPC and AI clusters that delivers high-bandwidth and low-latency for HPC, ML, and analytics applications.
Other Features
  • Remote extended network connections to the A*STAR, NUS, NTU, SUTD and NEA sites

 

  • Parallel file system (Lustre & PFSS)

 

  • Liquid cooled high-density Cray EX cabinets

 

  • Air cooled racks (specialized AI, Large memory, storage, login nodes)

 

  • Altair Workload Manager
Compared to ASPIRE 1, ASPIRE 2A provides
  • More than 3.5 times the processing cores

  • A 5-fold reduction in physical footprint (with 1.5 times fewer nodes)

  • Double the number of NVIDIA GPUs

  • Seven times the computational power
Prev
Next
Learn More

Interested to be a user of ASPIRE 2A+?

Find out more