Big news dropped recently: via Daring Fireball, we learn that Apple notified Imagination Technologies that they would no longer be using their products in new iPhone, iPad or iPod Touch designs within a 15 to 24 months timeframe.
For some time already, the GPU has been the biggest driver and bottleneck of iOS performance, if not since the beginning, at least starting with the iPad and Retina devices, compounded when iPads became Retina themselves: iOS SoCs have been characterized for some time as being bandwidth monsters (relatively to mobile devices), most of it connected to the GPU so that it can feed the screen pixels. It is the GPU which is mostly responsible for scrolling smoothness, for the amount of layers you can have on screen before performance takes a dive, for the performance of games, etc. The improvement of CPU performance, comparatively, improves the iOS experience much less (in the browser, mostly). If you’ve been curious enough to look at CPU teardowns of iPhones, for instance here for the iPhone 7, you know the GPU can take as much space as the multiple CPU cores, and for iPads a truly outrageous amount of silicon surface is taken by the GPU alone. And you are more than aware of Apple’s reliance on graphical effects (not just partial transparency, but also now translucency, blurs, etc.) in the iOS interface, all of which are generated by the GPU. So the GPU on iPhones and iPads has strategic importance.
If you need a refresher, Apple has been using PowerVR GPUs from Imagination ever since the original iPhone. More than that, though, it is the only outside technology (and a significant one, at that) that is and has always been an explicit dependency for iOS apps: readers of this blog don’t need to be reminded of Apple’s insistence to own every single aspect of the iOS platform (if you missed the previous episodes, most of it is in my iPhone shenanigans category) so as not to let anyone (Microsoft, Adobe, whoever) get leverage over them, but graphical technologies have been a notable exception, being more than mere software. For instance, while Apple uses OpenGL ES, and now Metal, to abstract away the GPU, a number of PowerVR-specific extensions have always been available and Apple encouraged their use. Even if Apple has recently tried to wean their developers away from these extensions, and stopped advertising to developers the GPUs as being PowerVR products (starting with the A7/iPhone 5S, if I recall correctly), iDevices are still using Imagination products, and PVRTC, as in PowerVR Texture Compression format, textures are still a common sight in the bundle of iOS games and other apps, for instance.
So the first challenge here is the dependencies on these extensions. I don’t see Apple getting developers to make such a transition so quickly, especially as the first devices without Imagination tech are going to be available 12 months before the deadline (the iOS product lines have become too complex to perform the hardware transition all at once), which would leave developers 3 to 12 months to transition… So most likely, Apple is going to have to support those, and this is going to expose them to intellectual properties issues (patents or otherwise). Besides the extensions developers explicitly use, there are all the performance aspects and tradeoffs specific to PowerVR that iOS games have unwittingly become dependent upon (e.g. whether to use complex geometry or compensate with shaders, how to best obtain some effects, etc.), which Apple would have to best reproduce, or at least not regress on, in a new GPU.
And even if they started from a blank slate when it came to third-party software, Apple has many technological challenges to overcome. Much like audio and video codecs, graphical processing technologies are patented to the hilt; but contrary to audio/video codecs, there is no FRAND licensing, no patent pool, or single licensing counter for GPU tech; instead, existing GPU companies live in an uneasy truce, given they are all exposed to each other’s patents. And mobile GPUs are a particular breed within this universe, with adapted techniques to live in such constraints, like Tile-Based Deferred Rendering (present in all PowerVR GPUs). Apple has managed to build its own CPU with great success, so I have little doubt that they will manage to develop their own GPU, especially given their expertise in SoC design as well. But I also see patent royalty payments in Apple’s future.
So what does this mean for iOS developers? For now, nothing. There is nothing to justify scrambling to remove any PowerVR dependency at this point, and it’s pointless to second-guess the performance characteristics of these future Apple GPUs. Best to wait for Apple to come forward. But there is some transition ahead, because at least some long-held assumptions about how iPhone graphics work are going to be challenged when the new Apple GPU will eventually appear. If anything, I’m surprised for such a glaring externality in the iOS platform to have managed to remain for so long, and it will be interesting to see how this will play out and how Apple will manage any necessary transition.
See also: Ryan Smith’s take at AnandTech, a reference.