ARM talks up fresh CPUs and a GPU, all tuned for AI
Cortex-A75, A55, and Mali-G72 coming next year
Chip designer ARM on Monday plans to announce its first set of processors based on its DynamIQ microprocessor architecture, in conjunction with a revised GPU chip.
DynamIQ, explained ARM product marketing head John Ronco in a phone interview with The Register, represents "a new way of combining CPUs together and giving more configuration options in how they're implemented."
The technology allows cores to be a variety of different sizes, said Ronco. This expansion on the big.LITTLE architecture that debuted in 2011 – in which cores are grouped, with size variations among groups – has potential benefits. For example, a chip with one large core and seven small cores could use the smaller cores for ongoing computation tasks and the large core for performance-intensive applications.
"DynamIQ is a new way of stitching cores together," Ronco explained. "It enables you to mix and match and have more possibilities in how you design. We're expecting to see more people use the big cores in the mid-range devices."
The first two CPUs to take advantage of this chip architecture are the ARM Cortex-A75 and Cortex-A55, which should begin to appear in hardware makers' devices during the first quarter of 2018.
The Cortex-A75 has been designed for performance and can be expected to serve as the basis for the SoCs in flagship mobile phone models, other computing devices, infrastructure, and automobiles. Running at 3GHz, it's more than 50 per cent faster than a Cortex-A73 running at 2.4GHz, based on SPECint2006 benchmarking specifications, according to ARM.
The Cortex-A55 is tuned for efficiency. "It's incredibly power and cost efficient," said Ronco, who expects it will be used in many mid-range phones.
Ronco noted that the A55's predecessor, the A53, has become the most widely deployed 64‑bit CPU ever.
ARM claims that the A55 in 16nm devices is 50 per cent more power efficient than the previous-generation A53 in 28nm devices.
"What matters a lot in phones is sustained performance because phones are very constrained in the power budget you can use," said Ronco.
ARM also plans to announce the Mali‑G72 graphics processing unit, successor to its Mali‑G71. The company pointed out that a billion of its GPUs were sold last year, about twice the merchandise moved in 2014.
The G72 has been optimized for graphically intensive mobile gaming, mobile VR, and on‑device machine learning, according to Anand Patel, director of product marketing.
According to Ronco, customer interest in AI and machine learning among customers is on the rise. "That's changing slightly the kind of compute tasks we have to do."
The latest Mali design includes optimizations that enhance machine learning computation. Its general matrix-to-matrix multiplication (GEMM) calculations are 17 per cent more energy-efficient, for example.
In a phone interview with The Register, Abhi Dugar, an analyst with consultancy IDC, said AI has become very important to ARM customers like Qualcomm. ®
Sponsored: Beyond the Data Frontier