News

Arm Ramps Up 5G Portfolio With New CPUs, GPUs, and NPUs

May 28, 2020 by Antonio Anzaldua Jr.

With an eye on 5G at the hardware level, Arm is attracting notice from Facebook, Snapchat, Google, and Netflix.

With 5G around the corner, Arm has enlisted its engineers and neural network programmers to up the capabilities of mobile devices by expanding their 5G-oriented IP for hardware designers.  

 

Arm's new 5G portfolio

Arm anticipates its 2020 mobile IP suite to be of use in autonomous, AI, XR, and smartphone applications. Image (modified) used courtesy of Arm

 

Arm says they provide the largest computing ecosystem for next-generation mobile solutions, specifically those that will operate in a 5G era. Earlier this week, Arm announced new 5G solutions that will help hardware developers to furnish a new "immersive" digital world.

 

An Upgraded Cortex-CPU for Mobile Devices

One addition to Arm's 5G portfolio is the Cortex-A78 CPU, a processor designed to bring a multi-day 5G experience that bridges the gap between mobile and laptop performance. Arm claims this CPU performs 20% better than its predecessor, allowing it to manage high workloads with greater machine learning (ML) performance. This feature may be useful for designers working with foldable smartphones and large-screen handheld devices. 

 

Arm’s Cortex-A78 efficiency performance compared to the Cortex-A77

Arm’s Cortex-A78 efficiency performance compared to the Cortex-A77. Image (modified) used courtesy of Arm
 

Arm claims that the Cortex-A78 is one of its highest-performing CPUs for mobile devices. The company sees this device as especially useful in digital immersion experiences that will poignantly benefit from 5G—namely AR and VR.

 

New Program Supports Customized Cortex Products

Another 5G solution from Arm is the Cortex-X Custom Program. This program will allow designers to customize and differentiate beyond standard Cortex products. 

 

Side-by-side specs of the Cortex-A78 vs. the Cortex-X1

Side-by-side specs of the Cortex-A78 vs. the Cortex-X1. Image (modified) used courtesy of Arm

 

The program comes with its first customizable CPU, the Cortex-X1, which Arm trumpets as having a 30% higher peak performance than Cortex-A77. The Cortex-X1 offers 5 instructions decoded per cycle and 8 MOPS per cycle. The device also supports 64 kB of L1, up to 1 MB of L2 cache, and 8 MB of L3.

 

The Mali-G78 Supports 5G Graphics

Arm's expanded portfolio also includes the latest Valhall-based GPU, the Mali-G78. The Mali-G78 is said to offer a 25% increase over predecessors in graphics and supports a line of 24 cores.

The Mali-G78 also provides designers with a performance advisor, a tool that provides quick detection of bottlenecks and real-time reports to enable continuous integration and faster workflow. 

 

A New Neural Processing Unit Leverages Machine Learning

Machine learning (ML) is essential for new augmented reality (AR)- and virtual reality (VR)-based smartphone applications and smart home hubs. As part of this announcement, Arm introduced the Ethos-N78 neural processing unit (NPU) to help deliver greater on-device ML capabilities and up to 25% more performance efficiency than other NPUs.

 

Ethos-N78 is designed to provide flexibility to SoC architects

Arm says the Ethos-N78 is designed to provide flexibility to SoC architects. Image (modified) used courtesy of Arm

 

ML requires NPUs to be highly flexible and adaptable to satisfy a wide variety of functions, such as face unlock and voice user interface. Along with Ethos-N78, developers will have access to a software stack with two flow options: an offline compilation flow for embedded devices or an online flow for Android neural network hardware. 

 

Arm Teams Up With Facebook, Netflix, and Google

With an approaching era of 5G comes a promise for greater speed, connectivity, reduced latency. Arm’s ML ecosystem offers CPUs that provide classification, object detection, high resolution, segmentation enhancements for advanced image processing, biometric sensing, and health diagnostics applications. 

Currently, Facebook and Arm are expanding an ML framework to exceed standard CPU capabilities. As AAC contributor Vanessa Samuel recently discussed in her article on TinyML, low latency is critical while running ML on edge devices.

Arm hopes its ecosystem will provide ML powerhouses like Facebook, Snapchat, Google Maps, and Netflix the same expected CPU characteristics while being mobile.

 


 

As 5G becomes more of a reality, what do you look for in a CPU, GPU, or NPU? Share your thoughts in the comments below.