An emerging application that scientists have been working on for years is researching the human brain (and brains of other organic beings.) The European Ecole Polytechnique Fédérale de Lausanne’s (EPFL) Blue Brain Project, a Swiss brain research initiative, selected HPE to build a next-generation supercomputer for modeling and simulation of the mammalian brain. The new supercomputer is called Blue Brain 5. Its charter is neuroscience simulations, analysis, and visualization.
The HPE system selected comes from the company’s SGI acquisition. The system pictured above is the HPE SGI 8600. Underpinning the system is an eclectic mix of technologies that allows the Blue Brain Project to perform multiple types of simulations within a single system.
Blue Brain Project’s HPE SGI 8600 Hardware
In terms of actual hardware, the HPE SGI 8600 that the Blue Brain 5 will use is a mix. There are a total of 372 compute nodes in a system expected to reach over 1PFLOP of performance. That is easily a Top500 supercomputer as of the June 2018 Top500 list.
The 372 compute nodes utilize Intel Xeon Gold 6140 and Xeon Phi 7230 processors. Intel Xeon Scalable CPUs and Xeon Phi CPUs can utilize features like AVX-512 along with standard x86 compute to run a wide array of applications. What is extremely interesting with the HPE SGI 8600 installation is the use of NVIDIA Tesla V100. The NVIDIA Tesla V100 is the current top-of-the-line GPU with TensorCore technology for deep learning applications. Researchers will have the best technology that Intel and NVIDIA have to offer at this time.
Instead of Omni-Path which one may expect from a Xeon Scalable and Xeon Phi supercomputer, the system will utilize Mellanox Infiniband for networking. A total of 4PB of storage for the system will be provided by DataDirect Networks. The storage system will have 50GB/s bandwidth and an 80GB/s Infinite Memory Engine flash buffer.