ARM has upped the supercomputer ante against rivals IBM, Intel and Nvidia with the announcement of Scalable Vector Extensions (SVE) technology.
Developed for the ARMv8-A architecture, the SVE technology is already set to be used for the Post-K supercomputer being built by Fujitsu for the RIKEN Advanced Institute for Computational Science in Japan.
Unveiled at the Hot Chips conference in the US, the technology supports vectors from 128-bit to 2048-bit. Shifting the vector calculation problem from software to hardware, SVE technology will be a scalable extension to the ARM instruction set.
Vector processors drove early supercomputers, but were then replaced by less expensive IBM RISC chips in the early 1990s. In today’s high-performance servers x86 processors are used, but this could be set to shift with the industry seeing a renewed reliance on vector processing.
The move my ARM to introduce this alternative chip architecture is a sign of the company’s plans to move deeper into the server, data centre and high-performance computing (HPC) space while offering something different to rivals.
The new chip design could soon be running the world’s most powerful supercomputer if the plans for the Post-K supercomputer comes to fruition. Fujitsu chose ARM in July 2016 for the supercomputer, shifting from the 2GHz Sun Sparc64 cores used in the K supercomputer. If the build goes smoothly, the new supercomputer could be capable of 1,000 petaflops.
The announcement at Hot Chips follows the recent acquisition of the UK chip maker by Japanese company SoftBank. With the price set at $32 billion, the acquisition hopes to sharpen ARM’s focus in the server and internet of things space.
