Last year Qualcomm started rolling out its first chips for Android phones that supported upgradeable GPU drivers to optimize performance, so now it's doing a similar thing for on-device AI and machine learning. Droid-Life points out that during Google I/O, Google and Qualcomm have announced updatable neural network API drivers, representing a new model that will roll out along with Android 12.
While NN API drivers have usually been updated along with major OS updates, now the companies say they can roll out quickly via Google Play Services. Even better, the updates will be available for older chipsets and multiple versions of Android.
In an I/O presentation about advancements in machine learning, Google developers said the NN API could boost performance as though the phone had two additional CPU cores, while using less power and creating less heat. Qualcomm pointed to Google Assistant and Google Maps as examples of applications that use on-device AI processing and can benefit from the tweaks. In the video, they also showed how on-device machine learning powers features like live captioning or automatic background replacement.
For developers, the benefit is having one API with regular updates that they can target across multiple versions, and if other chipset manufacturers join the program, then the reach will extend even more. For users, we should be able to get smaller apps that run better and don't drain the battery to do things like voice or face recognition, while also not uploading your data back to a server, impacting your privacy.