Next-gen mid-range phones to get major boost from new NPUs, GPU

Silicon designer Arm is responsible for the CPU and GPU designs seen in virtually every Android smartphone today, and the firm has just announced a few new designs for mid-range devices.

The most prominent announcement is the Mali-G57 GPU, marking the first time we see a mid-range GPU based on the company’s new Valhall architecture. This new architecture is also the foundation for the Mali-G77 flagship GPU.

Arm claims that the Mali-G57 delivers a 1.3x boost in “performance density” over its previous mid-range graphics silicon, the Mali-G52 GPU. It also touts 1.3x better energy efficiency over the earlier GPU. You can also expect support for foveated rendering in VR and 60% better machine-learning performance.

Furthermore, the chip designer says the Mali-G57 sports double the “texturing performance” compared to the previous GPU. Arm adds that the GPU is also capable of handling HDR graphics, physically based rendering, and volumetric effects. In other words, it’s shaping up to be a pretty compelling piece of kit in theory if you’re a mobile gamer on a budget.

The previous generation Mali-G52 GPU found its way into several mid-range Huawei phones, such as the HONOR 9X and Nova 5 series. But it’s unclear whether the US trade ban means HUAWEI will have to stick it out with older GPUs.

AI silicon for more devices

The chip designer also announced two new NPUs, in the form of the mid-range Ethos-N57 and low-end Ethos-N37. Arm is claiming a performance boost of over 200% compared to “many other NPUs,” although this likely depends on the specific workload and other factors. These new NPUs are tuned for super resolution, object detection, image classification, and speech translation.

We haven’t seen Arm’s previous, high-end NPU (Ethos-N77) in smartphones to the best of our knowledge, as the likes of Samsung, Google, HUAWEI, and Qualcomm have all developed their own AI silicon. Nevertheless, Arm is pushing the Ethos-N57 for mid-range phones and smart home hubs. Meanwhile, the Ethos-N37 is being positioned as ideal for low-end phones, smart cameras, and digital TV.

Either way, we’re seeing machine learning becoming more pervasive in our lives, from smart speakers and TV boxes to budget smartphones. And with offline inference more important than ever thanks to services like the new Google Assistant and Android 10’s Live Caption, it’s clear that improved AI silicon will be put to good use.