AI hardware news: Arm’s new chip designs and Nvidia-powered boards from AverMedia
The coming battle for chip supremacy at the edge continues as chip designer ARM Holdings expands its lineup of processors for edge devices that boost machine learning capabilities. Meanwhile, ARM’s designs are being used in Nvidia-powered boards from AverMedia in edge devices such as video cameras.
ARM this week said it is expanding its Cortex-M and Ethos-U, chip lines. Due out in a year, the processor designs will be the Cortex-M55 and the Ethos-U55 micro neural processing unit.
Both will be artificial intelligence-capable processors for Internet of Things endpoint devices, and both fit into Arm’s edge product strategy for devices with power constraints.
The Cortex-M55 semiconductors — Arm’s first using its “Helium” technology – will work in embedded devices. Helium accelerates vector calculations, enabling “small, low-power embedded systems to manage the compute challenges in many applications, such as audio devices, sensor hubs, keyword spotting, and voice command control, power electronics, communications, and still image processing,” according to Arm.
Using the kick from Helium, the company says, the Cortex-M55 can boost machine-learning performance fifteenfold compared to previous M chips. A 500 percent increase in signal-processing performance uplift is also promised. That kind of performance increase warms the hearts of many in the edge computing sector, which will thrive as onboard AI and ML capabilities are spread from the cloud to the factory floor.
The Ethos-U55 is something new among machine learning processors, according to ARM. The firm is calling it a microNPU, or neural processing unit. ARM already has a line of NPU designs to sell. The microNPU, however, has been designed so that it only works when paired with recent Cortex-M-based processors. Harnessed together, the duo will accelerate machine-learning performance about fivefold.
The Cortex-M has been good for ARM, which is a British company purchased by Japan’s Softbank Group in 2016. ARM reported an operating profit of $125m for the fourth quarter of 2019.
In other artificial intelligence hardware news, AverMedia debuted two credit card-sized artificial-intelligence carrier boards that support Nvidia’s Jetson modules. Being small, the boards are aimed at video analytics in retail and urban infrastructure uses.
The first board, the EN715-AA00, is designed for Nvidia’s Jetson Nano in industrial settings demanding compact design and high heat tolerance.
The second board, the EX715-AA00, supports the Jetson TX1/TX2/TX2i 4GB module. It’s operating temperature is even broader, ranging from 40 degrees C to 85 degrees C. It, too, is aimed at video analytics roles in retail and urban infrastructure.
Nvidia has been successfully pushing more AI applications to edge devices, bringing ARM along with it. The Jetson Nano boards are based on quad-core ARM Cortex-A57 processor designs.
Sign of more open-source edge to come from Equinix as it joins LF Edge
Comments