American chipset designer, Qualcomm, has added support for Google’s open source machine learning service, TensorFlow, to their up-and-coming flagship Snapdragon 835 System-on-Chip. Qualcomm explain that they have been working with Google in order to optimize TensorFlow to run on the Snapdragon 835’s integrated Hexagon 682 DSP (Digital Signal Processor), which will result in significant advantages in both performance and power efficiency. Developers can and have been using Google’s TensorFlow technology to add deep learning and artificial intelligence features to their applications, which can be desktop or mobile based, on the device or via the cloud – Google uses TensorFlow in the Google Photos and it’s one of the key components of the image recognition search function built into the service. The Qualcomm Snapdragon 835 features the latest version of Qualcomm’s DSP, the Hexagon 652, which is a specialized processing unit embedded into the overall chipset. With the right coding, the Hexagon 652 can elegantly and efficiently handle certain tasks that would otherwise be handled by the CPU or GPU and at the source link, Qualcomm have included a video demonstrating how the Hexagon 652 speeds up and increases the accuracy of image recognition, and you can check it out by clicking on the source link down below.
In this video, Qualcomm demonstrate how an image recognition application simply works better when allowed to use the Hexagon DSP embedded in the Qualcomm Snapdragon 835 – although it will run using only the Snapdragon 835’s Kryo application cores. With the Hexagon DSP enabled, the application recognizes many more images in a given period of time and with a higher confidence level. If the application were to be used to recognize a set number of images, the DSP-enabled device would be processing data for much less time and so would use less energy as compared with the CPU. This means less battery used, less heat produced and a smoother, better experience for customers. In the real world away from a test bench, providing developers optimizes their software to work with Qualcomm’s System-on-Chips, applications will run more efficiently.
As Google’s TensorFlow is open source software, this means anybody is able to integrate the code into their own applications and services. Artificial intelligence, smart applications and services are likely to be a key trend for 2017’s smartphones and Qualcomm are carefully positioning the Snapdragon 835 at the heart of this. It’s likely that we will see a number of applications and services being designed for devices based around the Snapdragon 835 chipset, which is due to arrive in smartphones in the coming months.