On-device and cloud AI technologies complement each other, Huawei said, improving the experience for smartphone owners by providing more speed, more personalized data, and a better understanding of the world around it. The NPU will see traditionally cloud-based apps that use AI run in real time on the phone. While AI is most often associated with virtual assistants like Siri and Google Assistant, Huawei hasn’t launched an assistant of its own yet, but is slowly revealing more about how we’ll use the NPU.
Genuine benefits
To get a better idea of how the NPU inside the Kirin 970 will benefit you, Digital Trends spoke to Christophe Coutelle, Huawei’s vice president of software marketing, who revealed just how much of a difference that extra processor will make. Having hardware dedicated to AI means other resources aren’t kept busy dealing with a notoriously power-hungry process like AI, he explained, using the camera app Prisma as an example.
“For an app like Prisma, the phone uses the camera and therefore the ISP, plus the CPU, the GPU for display, and still needs capacity to run the AI model. Having hardware (in this case the NPU) dedicated to AI is guaranteed to speed up the process. This means a higher frame rate, smoother transitions, and reduced power consumption.”
The beautiful artistic effects in the Prisma app take time to create on the app, but are possible in real time through the camera with the NPU. The NPU brings the AI processing onto the phone, rather than it being solely dealt with in the cloud. Coutelle explained why this is important, using another popular app to make his point:
“A lot of software that needs cloud-based services, like Pokémon Go (again, like Prisma, a highly power and resource-heavy app) will be improved and accelerated by the dedicated NPU hardware. It’s more power efficient.” Relying on cloud AI processing alone is like, “trying to go off-roading in a Formula One car. It’s just not efficient,” he joked.
This efficiency means the Kirin 970 will bring other desirable benefits with it. “With on-device AI processing you don’t have to worry about latency or radio power consumption. The user experience is much better. Privacy is also improved, as data is stored on the device. It’s not network dependent either — if you’re offline, these features still keep working. We also expect apps using AI running on a Kirin 970 device to consume less battery power than on those without the dedicated hardware.”
Better battery life when using power-intensive AI apps, which are only going to become more plentiful, while enjoying a better, faster, and enjoyable user experience? The NPU looks set to bring with it some genuinely exciting improvements that translate into benefits we all understand.
Image recognition
At launch, Huawei demonstrated the NPU’s super speedy image recognition. The Kirin 970 can process 2,000 images per minute, considerably more than other chips with similar capabilities, and it’s all thanks to the NPU. We’ve seen a demonstration of its image recognition skills too, and we were impressed with how quickly it could identify simple objects from chairs to birds. It also reacted at the same speed when faced with an unexpected image given to it by us. When shown a thumb, it didn’t know what appendage it was seeing, but did know it was attached to a person.
Additionally, the image recognition will understand what’s in front of the camera, and adjust settings and modes automatically. The NPU will promote development of innovative apps that use on-device AI, says Huawei; but developers have the choice to use either on-device or cloud AI, depending on which option best suits the app.
Technical specifications
What gives the Kirin 970 its power? It’s built using a 10nm process, like Qualcomm’s Snapdragon 835, and has an astonishing 5.5 million transistors in a space measuring a single square centimeter. Moving on from the quad-core Kirin 960, which powers the Huawei Mate 9 and Huawei P10, the Kirin 970 has an eight core CPU and a 12-core GPU. Huawei said it has 25 times the power and 50 times the efficiency of a quad-core chip with a Cortex A73 CPU, as used in the Kirin 960. Its power demands are low, usually pulling less than 0.5 watts, which means the phone won’t run too hot, or require special management of the processes to keep temperatures under control.
Huawei has big plans for artificial intelligence, but hasn’t leapt straight in with a virtual assistant, so where will the Kirin 970 and the NPU be used, and what are the precise benefits? More will be revealed in October when the company will announce the Mate 10, the first phone to use the Kirin 970.
Update: We spoke to Huawei’s Christophe Coutelle to find out more about the NPU’s benefits.
