Imagination announces new PowerVR GPUs, NNAs

Imagination Technologies has unveiled its latest PowerVR Series9 graphics processors, alongside a neural network accelerator (NNA) it has dubbed the Series3NX.

Still recovering from the loss of its biggest customer, Apple, Imagination Technologies hasn't given up on the embedded and mobile graphics market: The company's new Series9 graphics processing intellectual properties (IPs) are claimed to be considerably more efficient than their predecessors while adding in new features - and, interestingly, spanning two architectural generations: Rogue for the low-power end and Furian for the high-performance end.

'With PowerVR GPUs, we are focused on helping our customers create products that stand out at any price point,' claims Nigel Leeder, executive vice president for PowerVR at Imagination. 'These new GPUs are the best-ever mainstream PowerVR GPU cores and will enable us to continue to increase our Android device footprint and our overall volume across mobile, set-top boxes, DTVs, and automotive applications. With these latest GPUs, our customers can deliver even more differentiated and exciting products from entry-level to high-end.'

The new family starts with the PowerVR Series9XEP, Rogue-based and implementing the PVRIC4 visually-lossless compression algorithm for reduced bandwidth and memory footprint. This, Imagination explains, is primarily targeted at set-top boxes, digital TVs, lower-end smartphones and tablets, and mid-range in-vehicle dashboards and entertainment systems.

The PowerVR Series9XMP retains the Rogue architecture but, the company claims, improves the performance - making it only natural that Imagination is targeting mid-range mobile devices alongside higher-end set-top boxes and digital TVs. Interestingly, Imagination is also positioning the Series9XMP at devices which need to run artificial intelligence (AI) and other neural network algorithms, claiming its GPU can accelerate their performance considerably over the competition.

The PowerVR Series9XTP rounds out the range, switching to a revision of the company's Furian architecture which offers the best performance of the family - though, Imagination claims, it comes with optimisations to keep a balance between power, performance, and area. The Series9XTP is targeted firmly at high-end mobile devices, though the company is also sniffing around other markets including the data centre.

For those looking to do neural network tasks, any one of the GPU designs can be paired with Imagination's new Series3NX neural network accelerator (NNA), which is claimed to offer between 0.6 to 10 trillion operations per second (TOPS) in a single core depending on your desired power draw, while multicore variants can scale beyond 160 TOPS. Compared to its previous releases, Imagination is pointing to a 40 percent performance boost for the same silicon area, a near-60 percent boost in performance-versus-efficiency, and a 35 percent reduction in bandwidth requirements.

'The Series3NX architecture and Series3NX-F [Flexible] are built without compromise,' claims Imagination's vice president for vision and AI Russell James. 'Together they bring flexibility and scalability, while near doubling top-line performance. This is a game changer, a true enabler for mass AI adoption in embedded devices.'

Thus far, neither Imagination nor its customers have come forward to suggest when the new parts will appear in consumer devices.