Supermicro’s HPC System Management

HPC system management

Supermicro takes pride in developing high-performance computing (HPC) system management solutions from scratch. This establishment has a team that optimizes every solution’s aspects, from design to implementation, thus meeting and exceeding the significant difficulties in HPC system management. They simplify the implementation of supercomputers.

HPC System Management Solutions

Supermicro’s contemporary architecture has been designed to satisfy everyone’s HPC requirements. This form factor has several advantages, including multiple building blocks from system configuration to liquid cooling structure and a wholly integrated rack and motherboard design. It focuses on offering solutions that match the specific needs of its clients.

Enterprise High-Performance Computing Reference Design

Supermicro developed this flexible reference architecture for all-purpose, computing-intensive workloads. It is perfect for various use cases of enterprise HPC system management, including financial hazard management, digital manufacturing, gas and oil exploration, and medical research. Its specifications include:

  • SuperBlade® SBI-420P-1C2N servers, which have around 80 nodes per rack
  • Up to 6400 processor cores (Intel® Xeon®), 160 2.5-inch drive bays per rack
  • Fewer cables with built-in IB HDR and 25 G switch modules
  • Perfect serviceability and contain hot-pluggable nodes (PSU, FAN modules, and switch modules)
  • It is highly scalable, with 800 compute interchanges per cluster and ten racks

Reference Design: Scientific Research Supercomputer

This is a proven high-performance computing design developed for multiple research projects by scientists. This supercomputer can quickly adapt to weather simulation, gene sequencing, space and earth discovery, and nuclear reaction pure science simulation, among other simulations. Its specifications are:

  • BigTwin® AS-2124BT-HTR or 16 TwinPro® SYS-220TP-HTTR servers with 64 interchanges per rack
  • Around 5120 cores of Intel Xeon to a rack (TwinPro) or 8192 cores of AMD EPYCTM to a rack
  • DLC has liquid cooling (optional) and a 40KW CDU (capable of reaching 80KW)
  • About 30 racks and 1920 nodes make up each cluster within a 2:1 OPA structure
  • Low cost per HPC with a single socket of TwinPro per interchange design or unrivaled performance with a dual-socket per interchange BigTwin design

HPC Deep Learning, AI Reference Design

Supermicro’s AI + HPC plug-and-play device, pre-installed with the latest AI capabilities as they emerge, has unprecedented speed in discovering scientific research and enabling a research center to focus on developing new AI applications for processing natural language and autonomous driving. It is made up of the following:

  • AS -4124GO-NART or SYS-420GP-TNAR servers with NVLink® and NVIDIA® HGX A100, 5 4U 8-GPU structure
  • 80 NVIDIA A100 GPU systems with 40 petaflops of AI performance for each rack
  • Supports 30 hot-swappable NVMe drives per rack
  • GPU interconnects with high speed and 2 100G Ethernet portholes plus 8 200G HDR per node
  • Highly scalable, with 800 A100 GPUs and 100 nodes per cluster within a lock-free IB HDR structure and D2C liquid cooling

Liquid Cooling Solutions

Supermicro has superior liquid cooling solutions that have shown effectiveness in proven deployments and reduced environmental ramifications in HPC system management. Switching to liquid cooling has multiple benefits, such as energy savings, reduced air conditioner, fan operation, and increased efficiency. Future generations of GPU systems and CPUs may require liquid cooling because of their efficiency.

Liquid cooling solutions improve the PUE of GPUs, high-power central processing units, and high-performance data centers. It reduces the environmental impact and costs by lowering the carbon emissions from fuel plants.

Success Cases

Supermicro can customize HPC system management solutions to perform several workloads, huge capacity storage software used in different fields, or compute-intensive tasks. Supermicro’s HPC solutions could be tied up with other commercial applications and open-source platforms with evident success.

You May Also Like


About the Author: Micky Aron