ARM DynamIQ Chip Architecture Boasts 50x Uplift In AI Performance And Up To 8 Cores Per Cluster
But as we enter into an age where artificial intelligence (AI) is becoming a bigger part of our lives, we need even more powerful processors to leverage these capabilities. As a result, ARM is looking to advance its big.LITTLE architecture with a forward-looking design that will stretch into the next decade.
With that in mind, ARM is announcing its DynamIQ architecture, which it labels “Multicore Redefined”. DynamIQ allows for up to eight cores per cluster, with each core capable of being from different Cortex-A families. For example, 1+3, 2+4 or 1+7 big.LITTLE configurations could be deployed (or any configuration in between), which ARM says allows for its partners to create the “right processor for the right task” and gives more granular control over performance and thermal management.
Power management takes a step forward, as changes to each individual core’s clock, voltage, and sleep states can be changed independently (this should allow for the greatest savings in power/thermal budgets as processors conform to the task at hand), allowing for designs that scale from ultra-thin smartphones all the way up to high-end, enterprise-class servers. ARM has also redesigned the memory subsystem (shared across all cores) to not only improve power management, but to also to speed up data access.
Taken together, ARM says that the next generation of Cortex-A processors using DynamIQ will be capable of achieving a 50x increase in AI performance over the next five years (compared to a contemporary Cortex-A73 design). ARM also promises a 10x increase in communications speeds between the processor and onboard hardware accelerators. On that last point, ARM says that this performance boost will be a boon to computer vision and machine learning systems.
Lower latencies have also been implemented to ensure faster response times in decision making in mission critical fields, like autonomous driving where Advanced Driver Assistance Systems (ADAS) are increasingly finding their way into today’s vehicles.
Previous (and current) generation Cortex cores aren’t capable of supporting DynamIQ, which means that we’ll be seeing new core designs over the coming months which incorporate the architecture. And hopefully at that time, we’ll receive some hard data on how performance and overall efficiency compares to existing big.LITTLE configurations.
“ARM DynamIQ technology is the new foundation for smarter, faster, and more powerful user experiences for the next generation of intelligent devices,” says ARM’s Govind Wathan. “It will fuel innovation in an ever-demanding market for new designs, from thinner devices to cloud-based solutions – wherever computing happens.”
ARM hopes that leveraging DynamIQ architecture will allow its designs to be shipped in over 100 billion chips between now and the year 2021. To put that in perspective, it took ARM 22 years (1991-2013) to ship 50 billion chips, and four years (2013-2017) to ship it another 50 billion chips.