Intel Is Planning To Drop PowerVR Graphics
Going back to the original GMA500, a.k.a. the very notorious Poulsbo, Intel relied upon Imagination Technologies and their PowerVR SGX graphics core for their low-power Atom SoCs. Intel didn't use any in-house graphics core for Poulsbo and newer hardware like Moorestown and Medfield since they didn't have anything to compete in terms of great performance on an extremely low-power budget.
This has been a big mess especially for Linux users since Intel isn't allowed to provide an open-source driver (or any documentation) since they don't own control over the graphics IP and Imagination Technologies generally isn't an open-source friendly firm. PowerVR graphics have been a mess under Linux for not only Intel but also for the multiple other vendors relying upon the very common PowerVR technology within their hardware.
The good news for Intel customers is that it looks like Intel's planning to do away with this graphics IP in an upcoming SoC and replace it with their own in-house graphics core that can compete in the very-low-power segment while boasting respective OpenGL performance and other capabilities.
Intel didn't have the abilities back with Poulsbo to beat out the PowerVR SGX in terms of performance at the same power level, but it looks like they're finally hitting -- or exceeding -- their internal marks with future hardware. When it comes to original Intel graphics, Sandy Bridge is now really great (especially after a year of remarkable driver optimizations), Ivy Bridge can be as twice as fast as current-gen Sandy parts while not increasing the thermal envelope, and Haswell is sounding beautiful with the 2013 desktop parts potentially competing with mid-to-high-end discrete graphics processors.
Based upon information from different sources, it looks like Intel is planning to trim down a "Gen7" (Ivy Bridge) graphics core and lodging it on a future SoC. It would make sense that Intel would be looking to avoid its dependence on using Imagination's low-power graphics technology that's used by so many competing ARM SoC vendors.
If Intel can manage to deliver viable Intel-inside graphics within a near-future SoC across power, performance, and features, they may be able to attract more device vendors instead of ARM. With in-house graphics hopefully leveraging their existing and mature driver code-base, they would also be able to have an advantage on the driver side, especially if the support is available to everyone as open-source.
This work would also jive with what the Intel OTC desktop team has been doing on bettering the Mesa library support for Google's Android, advancing Tizen (the MeeGo successor) and its planned use of Wayland -- which necessitates a better driver than the current Poulsbo crap, etc.
The future SoC that's expected to introduce the new graphics is "Silvermont" and expected availability by early 2013. Before the exciting Silvermont there is expected to be the "Clover Trail" Atoms released this calendar year, but it doesn't sound like that SoC will bear new graphics capabilities. I hope to have some more details soon and am very excited to see Intel do away with the PowerVR graphics and its horrific driver support.