Simon Forrest explains how embedded chips can meet the challenge of delivering true local AI processing. GPUs and NNAs are rapidly becoming essential elements for AI on the edge. As companies begin to harness the potential of using neural networks
NEURAL NETWORK ACCELERATOR
The highest performance neural network inference accelerator
Neural networks are the established software tool for complex signal processing and pattern recognition that lie behind many AI technologies. PowerVR Series3NX is the fastest, most power-efficient embedded solution for hardware acceleration of neural networks on the market. Thanks to key architectural enhancements the Series3NX benefits from a 40% performance boost over the previous generation, performing up to 10 tera operations per second (TOPS) from a single core, providing the highest performance density per mm2 in the market.
Bringing multi-core scalability to the embedded AI market
Thanks to the scalability of PowerVR Series3NX architecture, multi-core implementations can achieve up to 160 TOPS, enabling ultra-high performance for the most demanding applications. Series3NX will be available in a variety of offerings, enabling SoC manufacturers to meet a range of design targets to address multiple markets and applications.
Neural network acceleration for edge devices
As neural networks drive an explosion in technological progress across industries, NNAs are now a fundamental class of processor, as significant as CPUs and GPUs. By integrating a PowerVR Series3NX Neural Network Accelerator (NNA) manufacturers can build devices that offer high-performance computation of neural networks at very low power consumption, in minimal silicon area. Offering this processing in edge devices removes the limitations of the cloud, such as bandwidth constraints, latency issues and privacy concerns.
Flexible bit-depth data type support
As a fully flexible solution, the Series3NX supports neural network bit depths from 16 down to 4-bit, reducing bandwidth and increasing performance without compromising inference accuracy.
Lossless weight compression
Complementing its low-bit depth support, the Series3NX introduces a new lossless weight compression scheme that reduces network model sizes and bandwidth thus increasing overall performance.
Advanced security enablement
PowerVR Series3NX integrates with the industry-leading security architectures including a flexible infrastructure that enables integration into custom solutions, enabling rights holders to protect their content where required.
Leading performance with low power consumption
With the industry’s highest inference/mW, the Series3NX delivers class-leading neural network acceleration with the lowest power consumption.
Introducing PowerVR Series3NX-F
The Series3NX-F brings programmable extensibility to the Series3NX architecture. It combines a Series3NX core with a neural network programmable unit (NNPU); a highly neural network optimised GPGPU, based on our industry-proven Rogue architecture. This provides developers with even greater flexibility in optimising their applications to run on the Series3NX architecture.
Putting the smart in smartphone
In mobile devices with a GPU, device manufacturers can pair a PowerVR Series9XE/XEP or 9XM/9XMP GPU with the Series3NX NNA in the same silicon footprint as a competing standalone Machine learning is now deployed in a wide variety of mobile applications, such as face recognition and verification, object recognition, image enhancement, style transfer and music tagging to name but a few. To support this, our Series3NX NNA cores deliver a paradigm shift in performance, while simultaneously reducing battery consumption over pure GPU solutions.
Security and surveillance
PowerVR Series3NX NNA cores enable a new class of smart camera that perform high-performance neural network-based analytics for a wide range of verticals such as commercial and home surveillance, retail analytics and drones. It supports classic use cases such as number/license plate recognition, person/object recognition, behaviour detection and perimeter defence.
Convolutional neural networks (CNNs) are playing a crucial role in developing self-driving cars. The Series3NX NNAs will power advanced driver-assistance systems (ADAS) including driver alertness monitoring, driver gaze tracking, seat occupancy, road-sign detection, drivable path analysis, road user detection and driver recognition.
Augmented and Virtual reality
Neural network hardware acceleration will be critical to fulfil the potential of next-gen augmented and virtual reality use cases. Scene understanding will enhance augmented reality, while movement analysis, eye tracking and gesture recognition will provide context awareness in virtual reality to provide the best possible relative user experiences.
An extensive ecosystem of tools and support
PowerVR GPU technology is driven by one of the world’s largest engineering teams dedicated to graphics processor development. It is complemented by Imagination’s PowerVR Insider ecosystem, which provides extensive support and tools to an extensive and vibrant community of developers, who have already created hundreds of thousands of apps optimised for PowerVR powered devices.
Design Optimisation Kits
To help our customers achieve the best possible implementation in the shortest possible time, we offer Design Optimisation Kits. They are a complete solution comprised of IP, libraries from partners, optimised reference floorplans and flows developed by us.
PowerVR SDK and Tools
Including a cross-platform OS and API abstraction layer, as well as a library of helper tools for maths and resource loading. It also features optimised example applications to demonstrate the most efficient ways of implementing common 3D graphics effects on PowerVR GPUs.
Our suite of utilities is designed to enable rapid graphics application development. It targets a range of areas, including asset exporting and optimisation, PC emulation, prototyping environments, online and off-line performance, analysis tools and more.
News from Imagination
China has been widely tipped to take the lead in artificial technology so when Imagination travelled to Shenzhen, Shanghai and Beijing in June this year to lead seminars on AI, they were well received by our industry-leading partners and customers
Recently, we ran a webinar entitled, “Enabling efficient implementation of neural networks in smart cameras”. If you missed it, it’s worth checking out, as we take a close look at the smart camera market and the need to embed neural
For many years, the semiconductor industry has strived towards tightly integrating more and more components into a single system-on-chip (SoC). After all, it is an entirely practical solution for high volume applications. By optimally positioning the various cores, memories and