Having dominated the mobile world for some time, ARM—the company whose CPU design is in Apple’s iPhone—is now going after the fastest computers in the world.
The chip-design company wants to graduate its processor architecture and take on complex calculations that drive tasks like weather modeling, economic forecasting and scientific research—the domain of high-performance computing (HPC).
Next month, ARM will detail the “ARMv8-A Next Generation Vector Architecture for HPC” at the Hot Chips conference. The architecture will be based on 64-bit CPUs, but ARM declined to share further information on it.
For ARM, HPC is a new market. While it has been successful in tablets and smartphones, it has failed to rock PCs, and it is struggling to get into servers, despite widespread interest on the part of hardware makers. There are software issues involved, and customers don’t want to change from x86 to ARM overnight.
ARM-based chips are built for mobile devices; their power-saving features give smartphones long battery life. But the ARM architecture will need an overhaul for faster systems, and vector processing features could push it up the computing food chain.
The new design could be something like that of Intel’s Xeon Phi supercomputer chip, which mixes Atom x86 CPUs and vector processors.
It’s also important for ARM to keep current supercomputing trends in mind as it formulates the new architecture. On its part, Intel has said it will include more machine learning features and instructions in future Xeon Phi chips.
Will the new chip architecture come to smartphones like the iPhone? Perhaps not anytime soon. But as the architecture evolves, ARM could adapt features for its mobile processor architecture.
Fujitsu has already adopted ARM processors for its next supercomputer, called Post-K, the successor to the SPARC-based K computer, which is ranked the world’s fifth fastest supercomputer by the Top500 list released last month. Fujitsu may use the new chip architecture being formulated by ARM.
Most supercomputers now run on x86 chips from Intel. IBM’s Power processors are also used in supercomputers. The world’s fastest supercomputer, TaihuLight, is however based on a homegrown Chinese chip.
In addition, coprocessors like Nvidia’s GPUs are also being used alongside CPUs for high-performance computing.