ARM Designs New CPUs, GPUs for Virtual Reality, Machine Learning

The Cortex-A75 and -A55 are based on the chip designer’s DynamIQ architecture, which offers greater flexibility, performance and energy efficiency.

artifical intelligence

ARM is targeting the growing artificial intelligence and machine learning spaces with its latest mobile processors, which are the first to be based on the chip designer’s new DynamIQ architecture.

As the annual Computex show gets underway in Taiwan, ARM is unveiling the Cortex-A75 and Cortex-A55 systems-on-a-chip (SoCs), which company officials said are designed to enable device makers to distribute intelligence farther out to the network edge and into the cloud.

At the same time, ARM is introducing its latest GPU, the Mali-G72, which is aimed at improving the performance of virtual reality and augmented reality, mobile gaming and machine learning applications.

Both of the Cortex chips come with instructions aimed at artificial intelligence workloads, and ARM officials said the chips will drive AI performance by 50 times over the current Cortex-A73 processors over the next three to five years.

The chips also come with more security capabilities, which officials said will be important going forward as the use of AI grows.

“Enabling secure and ubiquitous AI is a fundamental guiding design principle for ARM considering our technologies currently reach 70 percent of the global population,” Nandan Nayampally, vice president and general manager of ARM’s CPU Group, wrote in a post on the company blog.

“As such, ARM has a responsibility to rearchitect the compute experience for AI and other human-like compute experiences. To do this, we need to enable faster, more efficient and secure distributed intelligence between computing at the edge of the network and into the cloud,” Nayampally wrote.

At the core of the Cortex-A75 and Cortex-A55 SoCs is ARM’s new DynamIQ architecture, which the company first introduced in March and builds upon the big.LITTLE design, which was first announced in 2011.

The big.LITTLE architecture offered a higher-performance, more power-hungry larger processor and a smaller, more energy-efficient processor. The device’s operating system could send each task to the appropriate processor for improved performance or power efficiency, depending on the workload.

DynamIQ is designed to be more flexible, with ARM’s chip-making partners like Qualcomm being able to create clusters of big and little CPUs. The goal is to enable high performance while keeping down power consumption, extending battery life and reducing overall costs.

ARM also is stressing the importance of security in AI environments, stretching the company’s TrustZone technology to edge devices. Nayampally noted that a recent survey indicated that 85 percent of consumers are concerned about security and privacy related to AI.

The Cortex-A75 is designed for mobile devices like smartphones, laptops and other large-screen systems. The SoC delivers a 50 percent performance increase over its predecessor, and can also be used in such systems as self-driving vehicles.

The Cortex-A55 targets internet of things (IoT) devices and systems like gateways, which connect the IoT devices to the cloud. It will provide 2.5 times the performance-per-milliwatt of the current Cortex-A53 devices, according to company officials.

The Mali-G72 GPU is based on ARM’s Bifrost architecture and comes with enhancements that deliver 40 percent more performance on premium devices than its predecessor, which will drive VR and high-fidelity gaming applications.

It also brings 25 percent better energy efficiency and 20 percent better performance density. In addition, the GPU comes with optimizations and larger caches that will reduce bandwidth and provide a 17 percent efficiency improvement for machine learning workloads.

AI is driving a high level of competition in the chip market. Intel is making an aggressive push into the space by rolling out processors aimed at AI workloads and by buying companies like Nervana. Meanwhile Movidius to expanding its reach into such areas as field-programmable gate arrays (FPGAs) and autonomous vehicles. 

In addition, Nvidia has made AI and machine learning a key growth area for its GPUs, Advanced Micro Devices is doing the same with its Radeon GPUs and Google has its own Tensor Processing Unit.