ARM Community: ARM Mali Graphics, GPU Computing and reflections from shows - ARM Community

I’ve just about had a chance to recover from ARM’s recent triumphs, including at SIGGRAPH 2012: we were the first GPU IP vendor to gain OpenCL Full Profile conformance, we launched Mali-T624/Mali-T628/Mali-T678, we had ASTC adopted as a Khronos-approved extension, we helped launch OpenGL ES 3.0 and we showed off details of Transaction Elimination and our technical leadership in saving memory bandwidth and power. In addition to following the technical tracks at SIGGRAPH, I’ve also been to some other events, and I’ve had an opportunity to reflect on GPU Computing and our partners’ use of it. Here are a few recent thoughts:

Mobile is leading the rest of the industry. At the shows there were several moments that caused me to sit up and take notice and one was that everybody from the graphics industry is aware of the fact that the number of devices with really capable graphics is much higher in the post-PC world of new devices (mobiles, tablets, DTVs etc.) than the old markets of PCs and laptops, and they all want to get in on the act. Also, the gap between “high-end” and the new form-factors is becoming very blurred.

Midgard maturity. We demonstrated full customer release silicon of Mali-T604 (the first member of the Midgard architecture). We showed Full Profile OpenCL running on it and demonstrated the huge efficiency savings over computation on CPU as well as running demos showing the improvements in content that OpenGL ES 3.0 brings. The maturity of Midgard silicon seemed to surprise some people by comparison with others’ offerings.

Resolution wars. Screens are going to get higher and higher resolution, on all sorts of form-factor devices. There was a lot of talk about angles of subtending, visual acuity, dots per inch etc. but the summary is that people want super high-resolution screens on all their devices so they can’t make out the individual pixels.

Driving down nJ/pixel. The story for us is that our GPUs are going to have a push a lot more pixels. We knew this: we are already focused on this as we drive down energy per frame (drive down nJ/pixel) despite the increased complexity and resolution, through technologies like ASTC, Transaction Elimination and others yet to be announced. The story for other people (including developers) was that the first time they come across a device with a resolution like WQXGA (2560*1600), it could well be an ARM-based tablet and not a desktop PC.

GPU computing. In previous blogs, Tom Olson and I have sparred on the “killer use-case” for GPU computing. He said it was graphics and I said it was image processing, including computational photography. I am now, more than ever, convinced that we were both right. Some of our partners have committed plans to use OpenCL to implement the sort of camera functionality normally only found in DSLRs on mobiles and tablets (computational photography), but after listening to some of the leading-edge papers at the conference, it is clear that GPUs are starting to be used (even in the desktop) for the computation of certain graphical effects. And, wait for it, they are doing it because it is more efficient and it saves power. Yes, even in the 300 Watt PCI graphics card world, there is such a thing as too much heat and not enough power! The high-end is starting to worry about power a great deal, and of course I only looked a little bit smug, as at ARM, we’ve been designing for energy-efficiency for over twenty years…

Jem is an ARM Fellow and likes to think of himself as "The Godfather" to technical talent in ARM. After spending some time in his youth writing software for satellites and traffic-lights among other fascinating things, Jem spotted the technical inflection point of the mobile industry: graphics, video and other visual computing. As VP of technology in the Media Processing Division of ARM, Jem is busy with a lot of projects involving the future of cool ARM technology, which will revolutionise how people experience and interact with digital devices.