ARM launches microarchitecture to address AI

1 min read

Looking to meet increasing demand for artificial intelligence, machine learning, autonomous systems and virtual reality, ARM has unveiled DynamIQ, a new microarchitecture which it says will be the foundation for future Cortex-A processors.

Nandam Nayampally, general manager of ARM’s Compute Products Group, said: “DynamIQ technology is a monumental shift in multicore microarchitecture. DynamIQ technology will be pervasive in our cars, our homes and, of course, our smartphones, as well as in countless other connected devices where machine learning is applied.”

According to Nyampally, DynamIQ is an ‘evolutionary step forward’ for ARM’s big.LITTLE technology. Launched in 2011, big.LITTLE looked to address providing ‘the right processor for the right task’. DynamIQ enables configurations of processors in a single cluster which were previously not possible.

Up to eight cores can be designed into each DynamIQ cluster and each of these can have different characteristics. For example, it is now possible to create 1 plus 3 or 1 plus 7 DynamIQ big.LITTLE configurations and to provide more granular and optimal control. “This boosts innovation in SoCs designed with right-sized compute with heterogeneous processing that deliver meaningful AI performance at the device itself,” Nayampally continued.

Using new dedicated instructions, Cortex-A processors could see their AI performance increase by a factor of 50 in the next five years, compared to the performance available from Cortex-A73-based systems today. ARM also expects a tenfold decrease in response time between CPU and specialised accelerator hardware on the SoC.