Why the HUAWEI Mate 10 Pro’s NPU is the future of smartphone AI

Androids going rogue, all-powerful sentient networks causing global Armageddon – the picture painted hasn’t always been a positive one. With AI becoming a practical reality in recent years, however, the real-world outlook is, luckily, a lot more positive.

Why the HUAWEI Mate 10 Pro's NPU is the future of smartphone AI

Today, AI-powered algorithms can recognise your voice and respond in real time. They can translate foreign languages for you on the fly and they can help you organise, categorise and search through huge libraries of photographs without you having to laboriously tag them by hand.

And, amazingly enough, the power of the internet has brought all of this AI magic to the smartphone in your pocket, with digital assistants such as Amazon Alexa, Google Assistant and Siri, and automatic image recognition via Google Photos now available to pretty much every phone owner on the planet.

Cloud vs on-device AI

AI is so commonplace, in fact, that we already take it for granted. But there’s still a long way to go before it becomes truly ubiquitous and that’s because most AI computation today takes place in the cloud, on servers with dedicated AI hardware. In other words, you need an internet connection for it to work. Your phones and laptops aren’t as clever as you think; they’re merely conduits. The real AI resides in the cloud.

The processors in smartphones, although powerful, simply aren’t suited to the sort of repetitive computations required for things like voice and image recognition. Right now, not only would such operations take too long to complete on a phone, they’d hog resources, slowing the phone down in other areas, and they’d quickly drain your battery as well.

That’s going to change soon, though, and in the future, more phones will be like the HUAWEI Mate 10 Pro. The Mate 10’s Kirin 970 SoC (system on chip) processor includes a special module dedicated to AI calculations: the NPU (neural processing unit).

Without getting too deep into complicated mathematics, the idea is that the NPU is able to process the sorts of calculations involved with AI-type functions more swiftly and with less impact on the main CPU.


What can the NPU on the Mate 10 Pro be used for?

You might, at this point, be wondering what sorts of things this new silicon allows your phone to do. After all, today’s digital assistants work incredibly well and quickly, too. And how often are you really without an internet connection? Trouble is, they slow up when your internet connection does and disappear entirely when your connection goes down.

Phones with dedicated AI hardware could shift a lot of the work of your digital assistant to the phone and although you’d still need the internet for some questions, it could carry out other tasks – like launching an app, for instance, or playing a downloaded Spotify playlist – without the need for a data connection.

There’s so much more dedicated AI hardware can do than simple digital assistant work, however. This is ably demonstrated by the HUAWEI Mate 10 Pro and its most exciting party trick: live, on-the-fly scene and object recognition in the camera app. Most cameras, whether they be hulking DSLRs or smartphone cameras have a number of scene modes you can choose from, which prioritise certain settings that help the camera capture the shot you want. Select sports mode, for instance, and the camera favours high shutter speeds in low light, thus preventing motion blur. A food mode, meanwhile, is likely to accentuate, and slightly over-saturate, the colours.

On most cameras these modes are manually selected; not so on the HUAWEI Mate 10 Pro. Using its NPU to help it analyse the environment, the Mate 10 can recognise different objects and scenes, from text to food and faces, adjusting colour, contrast, brightness, and exposure in an instant. Not only that, but the Mate 10 also uses its NPU to help the camera select the perfect focus mode for the scene at hand.

The phone also has a specially accelerated version of Microsoft Translator preinstalled, which translates up to “500% faster than other smartphones running the same app” and without the need for an internet connection. It’s certainly nippy: point the Mate 10 at a French restaurant menu, take a picture and it translates it into English, without the need for an internet connection, in seconds.

AI and power management

But that’s not the end of the HUAWEI Mate 10 Pro’s AI talents. There’s also machine learning built into the core of the phone’s software, specifically its power management controls, which is great news if you’re sick of having to charge your phone every single day.

The Mate 10 Pro uses its NPU capabilities to analyse and monitor users’ habits and intelligently optimise its power management schemes. It can recognise which apps are important to the user and prioritise those, while preventing others from sapping precious battery life.

Not only that, but the phone is also loaded with a number of different profiles that allow it to spot different high-power drain scenarios – mobile gaming and camera use, for instance – and manage the phone’s resources in the interests of extending battery life.

Having the NPU and machine learning underpinning the power-management systems certainly seems to work. In fact, the HUAWEI Mate 10 Pro is among the best handsets you can buy for battery life right now, lasting up to two days between charges.

So far, there is no other smartphone with a dedicated AI processor, so the applications of the technology are still nascent and developers have yet to take full advantage. But with the Mate 10 Pro, Huawei is blazing a trail and offering a glimpse of an AI-powered future that isn’t tethered to the cloud.

It’s the sort of technology that – very soon – will spread into all aspects of our lives and, I’d wager, this time next year every phone and mobile chipset manufacturer will be following Huawei’s lead.

Disclaimer: Some pages on this site may include an affiliate link. This does not effect our editorial in any way.

Todays Highlights
How to See Google Search History
how to download photos from google photos