Qualcomm's Neural Processing Engine SDK to Enable Better Optimisation for AI

Artificial intelligence and neural networks have been at the core of almost every technological breakthrough that has come out in the last one year. Considering this, Qualcomm has now released its Neural Processing Engine (NPE) SDK that allows developers to accelerate the deep neural network or AI-type workloads on devices that have Snapdragon processors.

The Neural Processing Engine SDK, which is compatible with Snapdragon 600 and 800 Series processors, has been designed to support common deep learning frameworks including Caffe, Caffe2 and Tensorflow, and also offers support for custom layers, Qualcomm Technologies said in a press release. "The Snapdragon NPE is engineered to provide developers with software tools to accelerate deep neural network workloads on mobile and other edge devices powered by Snapdragon processors," the company said.

Essentially, with this developer kit, developers will be able to assign particular functionalities to different parts of the chipset. They will be able to allocate tasks to the CPU, GPU, or DSP to achieve the desired performance in any aspect. "Developers can take advantage of deep learning user experiences like style transfers and filters (augmented reality), scene detection, facial recognition, natural language understanding, object tracking and avoidance, gesturing, and text recognition to name a few," Qualcomm said.

Giving an example of the Facebook app, Qualcomm said that the company integrated NPE into the camera inside the app to accelerate Caffe2-powered augmented reality features. The company says that with the NPE, the app can achieve 5x better performance on the Adreno GPU, compared to a generic CPU implementation. "The result is a more fluid, seamless and realistic application of AR features when capturing photos and live videos," it said.

The Neural Processing Engine SDK includes runtime software, libraries, APIs, offline model conversion tools, sample code, documentation, and debugging and benchmarking tools, Qualcomm said.