As you know, the key feature of supercomputers is their high performance. Take, for example, the most powerful supercomputer currently available, the Summit, presented recently in the US. It consists of 9 thousand 22-nuclear processors IBM Power9 and more than 27 thousand graphics processors NVIDIA Tesla V100. However, the speed of data processing is by no means the only important thing in this matter. There is also energy efficiency, which the system can offer on ARM.
Last year, Hewlett Packard Enterprise (HPE) announced a prototype of The Machine, a supercomputer built on the basis of fast Memory-Driven Computing. Now HPE engineers turned the concept into a real product, so the system Astra appeared – the world’s largest supercomputer on the ARM architecture. The project is being developed by the US Department of Energy specifically for Sandia Labs (Sandia National Laboratories) as a new experimental platform for nuclear research.
Astra is equipped with processors Cavium ThunderX2, which provide a higher density of the layout in comparison with the comparable x86-system. In addition, this chipset accelerates the work with memory by 33%.
The supercomputer is built on the HPE Apollo system, which includes more than 145,000 processor cores in 2,592 dual-processor servers. 28-core ThunderX2 uses eight channels, which is two more than x86 chips. Each of the system’s processors has direct access to a large memory pool. This is a significant difference from the current traditional systems, where each chip works with its small amount of memory. Peak power of the supercomputer reaches 2.3 petaflops.
In Sandia Labs, Astra will be part of the Vanguard pilot program, which aims to find new technologies for managing US nuclear stockpiles. This will be a kind of pilot project that should prove the effectiveness of ARM-based servers before the government invests in a complete upgrade of the system.
For more stuff visit our site techverses.com and discover what you want.