Which is why I pay particularly close attention when a company 'disses' its competitors, because (whether it admits this or not) this means either:

• The company's marketing department is incompetent (unlikely, albeit not impossible, or)
• The company has a tenuous-at-best grasp on the leadership torch

For example, last week I pointed out to you that every time Intel's Atom group publicly burps (choose your favourite alternative bodily discharge analogy, if you prefer), ARM offers me an opportunity to be indoctrinated with its perspective. This tells me, no matter how much the company and its license partners will vehemently disagree once they see what I've written, that ARM is mightily worried about Atom.

Similarly, when Nvidia's CEO, Jen-Hsun Huang, asserts that his company is 'going to open a can of whoop-ass' on Intel, or when Nvidia's VP of content relations, Roy Taylor, claims:

Basically the CPU is dead. Yes, that processor you see advertised everywhere from Intel. Its run out of steam. The fact is that it no longer makes anything run faster. You don’t need a fast one anymore. This is why AMD is in trouble and its why Intel are panicking. They are panicking so much that they have started attacking us. This is because you do still [need] one chip to get faster and faster – the GPU. That GeForce chip. Yes honestly. No I am not making this up. You are my friends and so I am not selling you. This shit is just interesting as hell.

I giggle (c'mon, admit it, you did too). I also conclude (as I've long already strongly suspected) that Nvidia's very, very worried about its near-future relevance in the PC platform that constitutes the bulk of its current product shipments, revenues, and profits.

Looking only at today's fiscal situation, you might conclude that I've gone quite daft. Nvidia's net income is up 34% from this same time last year, after all, with first fiscal quarter 2009 revenues topping $1.1B USD. So what's the problem? Consider:

• Nvidia's market share growth over the past several years has largely come at the expense of ATI, who has conceded the ultra-high-end (low volume, but high profit margin) segment of the business to Nvidia's GeForce 9800 series and its Quadro equivalents, whose prior-generation HD 2000 series was largely across-the-board uncompetitive with Nvidia GPU counterparts, and whose presence in 'Intel CPU accounts' has (for, I think, understandable reasons) evaporated since ATI's acquisition by AMD.
• Conversely, the total available market size for discrete GPUs has long been in decline (in spite of an overall growing PC market), as leading-edge game content redirects to consoles, as mainstream gaming content is tailored to run adequately on high-volume (Intel, predominantly) integrated graphics, and as other mainstream applications for GPUs remain largely (and disappointingly) non-emergent. Windows Vista-only DX10 has, no surprise, been a huge disappointment, and is already 'old news'. And don't you think, as I do, that it's pretty pathetic how Nvidia is now resigned to stating that whereas games might appear on a console first, they'll later appear on a PC in an enhanced state? Just how many folks, I wonder, will re-buy an 'enhanced' title for their computer that they've already purchased for their Wii, Xbox 360 or PS3?

Don't get me wrong. The system partitioning balancing act between:

• Functions running hardwired on dedicated silicon (i.e. what GPUs used to be)
• Functions running in software on special-purpose parallel processing engines (i.e. what GPUs are now, by virtue of their shader architecture embrace), and
• Functions running in software on general-purpose CPUs, albeit with embedded function-specific acceleration engines (ie MMX, 3DNow!, SSE, etc)

has long fascinated me and led to ongoing editorial coverage. Witness my October 2005 feature article and online addendums on the subject, or do a generic web search for my 'GPGPU' content. Modern GPUs' massively parallel architectures are suitable for much more than processing polygons and pixels; consider, for example, their application in:

• Still and video image encoding, decoding, transcoding and other processing tasks
• Audio processing
• Encryption and decryption
• Database search acceleration, and
• Physics processing (notably highlighted by Nvidia's acquisition of AGEIA, specifically for its software, and coming on the heels of Intel's prior-fall purchase of Havok)

Yes, I know that Photoshop CS4 will reportedly harness GPU-accelerated algorithms to a much greater degree than is done today (something that, I might add, I predicted in print almost three years ago). And yes, I know that OS 10.6 will reportedly include OpenCL APIs that enable applications to easily tap into GPU acceleration resources; maybe this'll motivate Microsoft to do the same with 'Windows 7'.

But in Larrabee, Intel will soon have a parallel-processing engine to rival Nvidia's, with a massive number of single-die shader processors based on Atom x86 CPU cores. If that doesn't say 'eventual integration onto the CPU', I don't know what does. And nearer-term, once Larrabee is up and running, Nvidia discrete GPUs will increasingly find themselves shut out of Intel accounts, no matter what the courts might declare. AMD has similar long-term integration aspirations with Fusion, and nearer-term you gotta believe that AMD CPU customers are already garnering substantial fiscal motivation to go AMD/ATI for graphics, too. Sorry, Jen-Hsun, supercomputers aren't going to pick up the slack.

Nvidia's recent canoodling with miniscule-market-share Via has 'desperation' written all over it. And speaking of Via, the company is perhaps Nvidia's best chance for long-term relevance. With its ARM 11-based Tegra line (a fallout of Nvidia's purchase of PortalPlayer and, ironically, a revisit of a concept originally explored by a prior Nvidia acquisition, MediaQ), Nvidia has wisely concluded that the era of discrete multimedia 'engines' in mobile computing-and communications devices is rapidly drawing to a close, and has responded by launching highly integrated successors that go toe-to-toe with products from Marvell, Qualcomm, Samsung, TI and others. The company will need to do the same thing in the PC in order to survive there. Is Via's Centaur-based x86 license transportable to a purchasing agent? That's for the lawyers to decide. But it's either that route or an acquisition of AMD, which I personally find much less likely.

Tomorrow, I aspire, I'll follow up this writeup with a litany of interesting GPGPU applications I've collected over the past year. Until then, I welcome your thoughts on the admittedly controversial predictions I've just offered.