New HPE SGI 8600 14PF Jean Zay Platform Announced

HPE SGI 8600 Jean Zay
HPE SGI 8600 Jean Zay

Most supercomputer announcements happen in May or June before ISC or in November ahead of the Supercomputing conference. HPE has a new SGI 8600 based supercomputer announced this week. The new Jean Zay cluster HPE is building for GENCI is set to be France’s fastest supercomputer. The companies are using a traditional CPU and GPU architecture to advance both traditional HPC as well as AI research and development.

HPE SGI 8600 Jean Zay Supercomputer for GENCI

The new Jean Zay supercomputer is built atop of Intel Xeon Scalable processors along with NVIDIA Tesla V100 GPUs. With this system, GENCI and HPE are not using Mellanox Infiniband and instead selected to use Intel Omni-Path Architecture. They may be using built-in OPA like we saw in our Intel Xeon Gold 6148 Benchmarks and Review to significantly lower costs. GPU nodes using NVIDIA Tesla V100 GPUs are four GPU per system affairs which means that it is possible for the system to even use 1U form factors for the GPU nodes. The system is also using direct liquid cooling to lower cooling costs.

Here is an excerpt from HPE’s press release on the architecture:

Jean Zay will have a peak performance of 14 petaFLOPS and will be based on the HPE SGI 8600, a purpose-built high-performance computing (HPC) platform. It will converge the following powerful HPC and AI technologies to support machine learning and AI applications, along with improvement to traditional HPC workloads like modeling and simulation:

  • CPU and GPU Parallel Computing: To run various compute processors in parallel with up to four 100 Gbs links per node, we are using a single HPC interconnection network, the Intel Omni-Path Architecture (Intel OPA), powering 1,528 nodes of Intel Xeon Scalable processors for scalable compute performance and 261 nodes with 4 latest NVIDIA Tesla V100 32GB GPUs per node, for a total of 1044 GPUs.
  • Faster Simulation with Flash Storage Technology: We are enabling a read/write capacity of more than 300 GB per second to power faster simulations by integrating innovative flash storage capabilities with the HPE SGI 8600, an end-to-end HPC system, with support from DataDirect Network solutions.
  • Improved Rendering Time with Cooling DLC (Direct Liquid Cooling): Through an HPC-specific cooling solution, Jean Zay’s computing power is optimized to enable faster rendering times for simulations while also reducing energy consumption.

(Source: HPE)


Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.