Today, Imagination is launching its latest product: Imagination DXTP, a new GPU IP which extends battery life when accelerating graphics and AI workloads on smartphones and other power constrained devices. It is the final product of our most recent D-Series of GPUs and unites a range of features that have been introduced since 2022 into one solution, while simultaneously delivering a power efficiency boost of up to 20%. 

Discover the D-Series range in this blog. 

The smartphone market has moved on significantly since the launch of IMG DXT in 2022. The latest generation of foundation AI models has released a new wave of AI experiences that smartphone companies are integrating into their offering. From one-click photo fixing to real-time translations our mobile devices are smarter than ever before, and consumers are bringing forward their upgrades to access new features. 

However, phone fundamentals aren’t changing. They still retain a sleek form factor with a restricted battery, thermal limits are ever present, the user interface needs to remain responsive and there are household tasks that will always need running in the background. The extra performance required to deliver the benefits of AI to smartphones needs to fit into a phone on top of, and without affecting, existing requirements. 

Coming into this is DXTP, a processor that (like preceding Imagination GPUs) can handle graphics workloads, AI workloads, or both at the same time. The innate parallelism of Imagination’s GPUs make them equally well suited to handling the thousands of simultaneous operations within a neural network, as to the thousands of pixels for each frame in a game. It is a very flexible piece of silicon. 

But power efficiency is crucial in the mobile market. And this is exactly what DXTP brings – up to 20% power efficiency improvements simply from tuning a wide range of small blocks within the GPU, such as the texture processing unit, with a view to saving power. 

DXTP can also handle more complex graphics and increases sustained performance, this is all thanks to a slightly different arrangement of Universal Shading Clusters (USCs) within its Scalable Processing Units (SPUs) compared to other D-Series cores. By reducing the number of USCs per SPU from three to two (and then increasing the total number of SPUs from two to three to reach the same performance level), DXTP offers 50% more geometry throughput and better memory access, and the cache and system level bandwidth has also been made larger. 

On the compute side, DXTP features the same additional pipeline for FP16 operations as previous D-Series GPUs and boosts its compute performance further by avoiding setup bottlenecking on simple compute tasks by accelerating the scheduling of compute tasks up to 16x. It ships with support for Imagination’s highly optimised OpenCL compute libraries that boost GPU utilisation for common AI applications like computer vision, large language models and signal processing. These libraries work alongside GPGPU and AI inferencing reference kits. For Android developers, forthcoming LiteRT support for Imagination GPUs will help with the creation of high-performance mobile AI experiences.  

Developers can easily target DXTP through popular APIs like OpenGL ES, OpenCL and Vulkan. The DDK is regularly updated to support the latest versions of these APIs along with key extensions. The Imagination Developer Portal has the award-winning PowerVR SDK along with guides and sample code showing developers how to make the most out of Imagination GPUs. The Portal also hosts our feature-rich tools for low-level performance analysis, trace and debug which enable developers to achieve the highest possible performance level. 

DXTP is now the go-to GPU for pure graphics and AI acceleration on power-constrained devices. It has already been licensed for use in smartphones and automotive but is equally well suited to other consumer and desktop applications. If you’re interested in finding out more about DXTP, book into the Online Launch Event on Wednesday 5 March.