Imagination launches Neural Network Accelerator for AI applications

     Imagination Technologies has announced the launch of PowerVR Series3NX, its latest neural network accelerator (NNA) architecture for artificial intelligence (AI) applications. Building on its award-winning predecessor, the new Series3NX offers unmatched scalability, enabling system-on-a-chip (SoC) manufacturers to optimize computing power and performance for a range of embedded markets such as automotive, mobile devices, smart video surveillance, and iot edge devices.

     The performance of a single Series3NX kernel ranges from 0.6 to 10 trillion operations per second (TOPS), while its multi-core implementation scales to more than 160TOPS. Thanks to architectural enhancements including lossless weight compression, the Series3NX architecture delivers up to 40 percent more performance over the previous generation over the same chip area, enabling SoC manufacturers to improve performance efficiency by nearly 60 percent and reduce bandwidth requirements by 35 percent.

     As part of the Series3NX architecture, Imagination also released the PowerVR Series3NX-F (Flexible) semiconductor intellectual property (IP) configuration to deliver an unprecedented balance of functionality and flexibility, while also combining industry-leading performance. Series3NX-F customers can use the OpenCL framework to differentiate and add value to their products.

     "There is a huge opportunity to apply AI to the edge to create more powerful, autonomous and easy-to-use devices." Jeff Bier, founder of the Embedded Vision Alliance. "In many of these applications, a key challenge is to achieve the right combination of processing performance, flexibility, cost and power consumption. I applaud Imagination Technologies' continued investment in developing innovative processors to meet these needs."

     Russell James, Vice President of Vision and Artificial Intelligence at Imagination, said: "The Series3NX architecture and Series3NX-F are truly innovative products. Together, they bring flexibility and scalability, while nearly doubling the performance ceiling. This is a game changer that can really drive mass adoption of AI in embedded devices."

     To cater to this rapidly evolving market, the new PowerVR tools have been extended to optimally map emerging network models and provide the ideal combination of flexibility and performance optimization. By using Imagination's dedicated Deep Neural Network (DNN) API, developers can easily write AI applications for the Series3NX architecture as well as existing PowerVR Gpus. The API can work on multiple SoC configurations to make prototyping easy on existing devices.