Brian's Brain

Oct 12 2005 12:35PM | Permalink | Email this | Comments (0) |
Blog This! using:  Blogger.com | LiveJournal |
Digg This | Slashdot This | add to Del.icio.us

This blog post references my cover story, 'Instigating a Platform Tug Of War: Graphics Vendors Hunger for CPU Suppliers' Turf' in the October 13, 2005 edition of EDN.

Although the print article focuses the majority of its 'ink' on PC-centric applications for GPUs-beyond-graphics, please don't walk away thinking that's the only place where the concept makes sense. The PC focus was a necessity driven by non-infinite available print page count; I meant what I said at the beginning of the article, that "this tug of war isn't restricted to PCs; it will potentially play out in any system that includes a display". In fact, some non-PC scenarios are even more amenable to the idea; any time you've got an application with demanding performance expectations, an underpowered CPU and/or stringent energy consumption requirements, a hand-off from the system processor to more function-optimized silicon might make sense (also keeping in mind the relative bill of materials cost of the alternative approaches).

I'm reminded, for example, of Intel's early marketing efforts related to its ARM-based XScale processor line. The company attempted, in a thinly veiled effort to up-sell customers on faster and faster CPUs, to convince us that XScale (courtesy of Wireless MMX and other capabilities) could single-handedly handle all required 2- and 3-D graphics, and still and video imaging, functions, in conjunction with a low-tech LCD controller to tackle the final pixels-to-screen task. Those of us in the 'know' smiled and nodded, all the while winking at each other, and proceeded to advocate dedicated graphics-and-imaging silicon to our engineering audiences with even moderately-demanding visual processing needs. Companies like ATI and MediaQ (later acquired by Nvidia) didn't buy into Intel's hype, either. Lo and behold, Intel now sells the 2700G Multimedia Accelerator, and trophy platforms like Dell's Axim X51V showcase the advantage of a multi-chip multimedia architecture. And Falanx, Imagination Technologies (whose licensees include ARM, Freescale, Intel, Sony Ericsson and Texas Instruments) and Nvidia all sell shader-based GPUs for low power (i.e. handheld) applications.

But....there's always a 'but'. Remember that at the end of the print article's sidebar, 'The CPU Perspective', I point out that if the abundant rumours are true, Sony's Playstation 3 engineering team almost went with a CPU-centric graphics implementation. Here's another case study for you to chew on. Brian Schmidt, Microsoft's audio program manager, gave an excellent tutorial on audio-in-game console trends, understandably focused on the company's upcoming Xbox 360, at the Audio Engineering Society Convention last Saturday afternoon. The console makes significant sonic advancements over its Xbox predecessor in a number of areas; the number of simultaneous 'voices', the resolution and sample rate of each sound file, surround virtualization, reverb and other effects, etc. But whereas the audio algorithms (including real-time Dolby Digital encoding) ran, in the Xbox, on much-beloved and fondly remembered dedicated audio DSP circuitry within the console's Nvidia-developed core logic chipset, in the Xbox 360 it all runs exclusively in software across one or multiple of the console's three PowerPC CPUs.

And what about handhelds? Don't forget that at the recent Intel Developer Forum, Intel unveiled 'Monahans', a next-generation XScale CPU which is forecasted to run at up-to-1.25 GHz. The CPU-vs-GPU tug-of-war continues....I've said it before and I'll say it again....there isn't one implementation option that's optimum for every possible application. There is, however, one implementation option that's optimum for your specific application. Your job is to find it. Isn't engineering fun?