Servers – SGI 8600

The world’s most advanced production supercomputer


SGI8600
SGI8600

Clustered, High Performance Computing (HPC) system

Formerly the SGI ICE XA, the HPE SGI 8600 is a scalable, high-density clustered supercomputer system. Its liquid cooling translates into high power efficiency with substantial savings in energy usage.2 With industry-standard processors, interconnect, and software, SGI 8600 offers maximum flexibility and performance. Select from multiple interconnect and server tray options. This unique architecture is coupled with comprehensive system management and services.

Advanced interconnect technologies

HPE SGI 8600 consistently delivers high performance by integrating leading industry-standard interconnect technologies into clustered solutions. Add to this industry-standard software support and close working relationships with interconnect vendors.

Performance with innovative engineering

Building on the proven SGI ICE XA architecture, we leverage a collection of engineering innovations to enable the HPE SGI 8600 to deliver the best possible performance for an industry-standard clustered High Performance Computing (HPC) system.3 These include the most powerful CPUs, fastest memory, latest GPU technologies, highest speed interconnect topologies and tuning of the Message Passing Interface (MPI). Designed to scale to thousands of nodes, the HPE SGI 8600 also is well optimized for deep learning through NVIDIA SXM2 compute trays.

SGI 8600 Data Sheets:






Server Tray Portfolio

SGI8600HPE SGI 8600 with HPE XA730i Gen10 Server Tray

A quad node Intel® Xeon® Scalable Processor Family compute tray with 2 CPU sockets per node (total 8 CPU sockets). This model delivers maximum Intel® Xeon® Processor Scalable family compute density.

HPE SGI 8600 with HPE XA780i Gen10 Server Tray

A single two CPU socket node Intel® Xeon®Scalable Processor Family compute tray with support for up to four NVIDIA Tesla SXM2 GPUs with NVLink.

HPE SGI 8600 with HPE XA760i Gen10 Server Tray

A quad node Intel® Xeon® Phi processor compute tray with one CPU socket per node. All trays utilize an innovative cold sink technology to provide high node density at proper cooling levels.

Technical Overview

Density

  • Compact E-Cell design with up to 36 blades and 144 nodes per rack
  • 288 Intel® Xeon® processor sockets per rack
  • Integrated dual plane switching, power and cooling

·Efficiency

  • Direct attached liquid cooling
  • 100% heat removal via data center water
  • No heat transfer to data center room

·Scale

  • Scaling without use of any external switches
  • Multiple installations scaling to thousands of nodes