Apple drops Imagination Technologies: on the wrong side of a revolution

As the main supplier of GPU cores to Apple’s quickly growing business of smartphones, Imagination could only profit from its technology becoming the cornerstone of a new direction in computing. Whereas nVidia focused its work with the similar, and more proprietary, CUDA on desktop and server computers, OpenCL promised to provide something more mobile friendly. It would let GPUs give stressed general-purpose processors such as those from ARM a helping hand on arithmetic-intensive problems in both mobile and desktop machines.

Fast-forward almost ten years and it is Imagination that has suffered the financial damage. Two-thirds of the value of the company’s shares were wiped out in seconds as soon as trading on the London Stock Exchange got underway on Monday morning (3 April). Why? OpenCL is now a reasonably well-accepted standard, but Apple decided that in terms of the hardware it uses inside its phones and tablets it’s time to move on and design its own GPU.

The most obvious problem for Imagination is royalties from Apple for the existing range of phones and tablets accounts for half its total revenues and most of that is pure profit. If you take out the MIPS processor and radio cores it becomes clear that the OpenCL dividend has not paid off for Imagination over the past decade. Only about a third of Imagination’s GPU business is now outside Apple. Two-thirds of its market capitalisation suddenly vaporised.

In reality, it’s not as if Apple will never pay Imagination any money ever again. The computer company is still making phones and tablets based on processors it designed years ago. The new products that will be PowerVR-free will not appear for about another 18 months, it seems. There might be a smaller contribution in the years to come rather than no money at all even if Apple does its own GPU.

Imagination’s statement to shareholders implied Apple will need to work out a deal to pay the company for IP on the basis that it will be hard for the computer maker to avoid infringing its GPU patents. But the same argument could probably be made by AMD and nVidia who have tended to the thicket of patents that surrounds GPU technology. I haven’t forgotten ARM by the way. However, with Apple likely to continue to use the company’s general-purpose architecture for years to come, it’s difficult to see ARM management pushing too hard on that idea.

The problem for Imagination and, to some extent nearby ARM, is that companies like Apple now have enough experience and willpower to do more of the design themselves. For the past 20 years, the overwhelming trend in digital electronics has been to outsource pretty much everything that isn’t nailed down. First, they divested their manufacturing plants. Then companies turned to design houses and suppliers of intellectual property (IP) cores to do more of the engineering. Analysts talked of “design-less” companies that were pretty much branding and marketing operations that would just slap a badge on things designed and made elsewhere. The chances are your internet router comes from one of them.

Apple, though, has been swimming the other way. And it’s not alone. Facebook and Google in the US and Baidu in China have gradually been getting much more involved in hardware design. It’s not because it’s cheaper. It probably isn’t. Adam Smith’s advice on specialisation does still mostly work. But computing is changing. Why? The core reason is because batteries aren’t great at storing energy and digital systems have become too conspicuous at consuming more and more of the world’s electricity.

For the companies that have replaced IBM at the top of the tree, they now have an opportunity to use this to shape how the technology evolves. By doing more in-house they can control how the technology disseminates. They also have the opportunity to move faster than the pace at their suppliers who have focused on tweaking what they have rather than embarking on dramatically new architectures.

New applications such as machine learning have been shoehorned into GPUs because those GPUs happen to crunch more through floating-point calculations than any other form of processor. The GPU is far from ideal for the job, which is why companies such as Graphcore and Wave Computing have appeared. These architectures take much more interest in the memory surrounding the processor rather than the arithmetic pipeline itself because data transfers to and from memory are responsible for most of the wasted energy in digital computers.

Even the graphics applications have changed. To be fair, Imagination as well as ARM have embraced techniques such as ray-tracing to support virtual reality. However, the third-party IP suppliers have offered what amount to fairly small tweaks to existing architectures.

Although ARM is better protected than Imagination in this shift, its DynamIQ proposal, which has been aimed at machine learning and virtual-reality processing, really only puts more ARM processors in the box with a more streamlined memory interconnect.

Bringing development in-house provides Apple with the opportunity to make major changes. Its hardware will probably not look all that much like a conventional GPU – that would be a wasted opportunity at this juncture when the potential exists to provide something that can handle the fusion of machine learning, virtual reality and communications.

There is naturally a risk for a company retaining more of its architectural development in-house, as we saw with IBM’s fall from dominance. For the currently dominant IP suppliers they risk being left behind.