Qualcomm announces AI Engine for Snapdragon 845, 835 and more

Pixel Camera

Qualcomm is kicking off what is expected to be an AI-centric Mobile World Congress with a large AI announcement of its own.

The chipset manufacturer today announced the Qualcomm Artificial Intelligence Engine, a package of hardware and software elements that it hopes will facilitate many more AI-enabled experiences on handsets running Snapdragon processors.

In its announcement, Qualcomm said it’s already partnering with many smartphone manufacturers on using AI Engine components, including OnePlus, Motorola, Asus and ZTE.

Further, it reported that some of its partners are planning to debut AI features using the new AI Engine on future Snapdragon-powered flagships.

Director of product management Gary Brotman said in a briefing that some of the mobile AI applications consumers can expect to see might include on-device smart albums, single-camera bokeh (like the Pixel 2’s tech) and face authentication.

The AI Engine will be supported by Qualcomm’s Snapdragon 845, 835, 820 and 660 mobile chipsets.

The company specifically singled out the 845 for its on-device AI processing.

On-device processing allows for users to utilize AI experiences with or without a network connection, providing improved privacy and reliability.

The chipset uses its Hexagon Vector Processor, Adreno GPU and Kryo CPU to run AI on-device. Developers and manufacturers can choose the optimal core for the desired AI user experience.

The main software components of the AI Engine include a Snapdragon Neural Processing Engine software framework, which allows for the optimization mentioned above. It supports Google’s Tensorflow framework, as well as Caffe, Caffe 2 and the Open Neural Network Exchange.

The AI Engine also includes support for the Android Neural Networks API, first released in Android Oreo, and introduces Hexagon Neural Network library. The latter helps developers run AI algorithms directly on the Hexagon Vector Processor.

AI features have been a key selling point for many recent flagships, including Apple’s iPhone X, Google’s Pixel 2 and Huawei’s Mate 10.