The GPU of everything

4 mins read

At the end of 2019 Imagination Technologies made one of its biggest announcements in years with the launch of a new GPU architecture. The A-Series is, according to the company, its most important GPU announcement in over 15 years and it claimed that the series has the world’s fastest GPU IP.

The A–Series is an evolution of the company’s successful PowerVR graphics architecture and has been developed to meet the graphics and compute needs of a broad spectrum of next-generation devices.

According to Imagination it is the “GPU of Everything” as it is intended to address multiple markets from automotive, AIoT, and computing through to DTV/STB/OTT, mobile and the server market.

The announcement is certainly important for Imagination as the company is looking to show that the architecture is better than the custom GPUs currently being made by the likes of Apple and Qualcomm, and it’s seen by many as a bid by the company to make licensed GPUs more competitive in what has been a tough market for IP companies like Imagination and Arm.

As Dr. Ron Black, CEO, Imagination Technologies, said at the launch of the GPU: “The A-Series is our most important GPU launch since we delivered the first mobile PowerVR GPU. It offers the best performance over sustained time periods and at low power budgets across all markets.”

The launch follows a turbulent few years for Imagination and while the UK-based company remains a leading player in the mobile GPU space, licensing its GPU technologies, its PowerVR-branded GPUs are found in far fewer products than in the recent past having, in many cases, lost out to competitors such as Qualcomm.

In the past its high-end mobile GPUs had been used extensively by Apple in the company’s A series SoCs, up until the Apple A10 in 2016. In the low to mid-range, Imagination’s GPUs were also used by companies including MediaTek and Rockchip.

In a serious blow – at the time – for Imagination, its close relationship with Apple appeared to have ended when the tech giant announced that it was set to develop its own custom GPUs, replacing Imagination’s, and in a matter of months of the news breaking Imagination was put up for sale and bought by a Californian equity firm Canyon Ridge.

The relationship between Apple and Imagination has been renewed, however, with the signing of a new multi-year, multi-use license agreement at the turn of the year under which Apple will have access to a wide range of Imagination’s IP.

The strained relationship between the two companies had left Imagination without a key customer for its high-end GPUs at a time when it also found itself coming under increased competitive pressures.

In fact, by the middle of last year, the licensed IP from companies such as Arm and Imagination were being described as being significantly inferior to the custom IP that was then appearing from companies like Apple and Qualcomm.

That custom technology is unlikely to be licensed to third-party SoC vendors so that trend was starting to cause real problems for the wider industry.

Both Arm and Imagination responded with the launch of new products. The former with the Mali-G77 with the new Valhall GPU architecture; while Imagination launched the A-Series.

The A-Series has certainly delivered significant improvements, at the same clock and process, offering 2.5x the performance, 8x faster machine learning processing and 60% lower power than current PowerVR devices that are being shipped, according to Imagination.
Commenting Jon Peddie, principle and founder of Jon Peddie Research, said: “The simple fact is that for mobile SoCs the market leader owns its own GPU technology and is increasing market share at rate of 5% year-on-year. In order to stop the losses to their own potential share the other mobile SoC companies need a compelling GPU that will deliver some real competition.”

His point being that if licensed IP had continued to remain inferior to custom IP, then the wider market would have become increasingly uncompetitive.

The A-Series architecture is Imagination’s 10th generation GPU architecture and consists of three categories of GPUs: the XE, XM, and XT product lines, which represent the low-end, mid-range and high-end segments.

The A-Series architecture has four configurations in the XT series and one for the XM configuration. The AXT-64-2048 is a flagship GPU with 2.0 TFLOPS, 64 gigapixel/s fill rate, and 8 TOPS of AI performance; the AXT-32-1024 is described as a premium mobile GPU with 1 TFLOPS, 32 Gigapixel/s fill rate, and 4 TOPS of AI performance. The remaining configurations are the AXT-48-1536 and the AXT-16-512, while the AXM series has the AXM-8-256, which is rated for 256 GFLOPS, 8 gigapixel/s fill rate, and 1 TOP of AI performance.

Delivering for gaming
In part the A-Series has been developed to meet the needs of the gaming market.

“We have been working hard to develop this next generation of GPUs in terms of both the architecture and performance and I think we have created a GPU that our customers have been looking for,” said Graham Deacon, VP of Marketing – PowerVR Graphics IP at Imagination.

“With the A-Series we have focused on delivering a quality experience for gamers, especially when 47 per cent of the gaming market is now focused on the mobile phone.

“The A-Series has been about delivering a traditional gaming experience - we didn’t want the user to lose that experience - but via a mobile platform and at low power."

According to Deacon, most mobile GPUs struggle to deliver consistent gaming experiences due to thermal constraints – they are fast for a short while, then slow and then never recover, resulting in a disappointing user experience.

The A-Series looks to deliver a more sustained performance for extended game play at consistent frame rates avoiding thermal panic clock throttling, glitches or overheating.

“The A-Series’ PowerVR architecture benefits from an inherent bandwidth and power advantage because its tile-based deferred rendering technology means it only draws what is visible on screen. This is combined with advanced direct flexible power signalling using Pro-Active DVFS (Dynamic Voltage and Frequency Scaling) and Deadline Scheduling algorithms, which deliver fast power control with an ultra-low latency response. If parts of the GPU aren’t fully utilised or needed for processing, they are immediately slowed down or even put to sleep to ensure optimal power efficiency,” Deacon explained.

To deliver this high power efficiency is necessary and is provided by PowerVR’s smart flexible firmware based low-latency DVFS algorithms with a direct GPIO control path into the system.

The A-Series also features Imagination’s HyperLane technology that comprises of individual hardware control lanes, each isolated in memory, enabling different tasks to be submitted to the GPU simultaneously for secure multitasking.

With this Dynamic Performance Control, the GPU is able to spread its performance across multiple tasks, executing them all, while maximising GPU utilisation. Priority HyperLanes deliver a dynamic split; for example, prioritising graphics and delivering all the required performance for that application while an AI task could then be run alongside using the remaining performance.

In the coming year Imagination is planning to reveal a new B series architecture which will include 30% performance improvements, which in turn will be succeeded with a C and D series in 2021 and 2022, again with 30% performance improvements being promised.
Imagination’s announcement of the A-Series architecture has been welcomed and been described as providing a healthy dose of new competition into the mobile GPU market.