CPU Wars: Nvidia Suggests It's Working On An x86 SoC

The discrete GPU market is in free-fall, with Intel’s offering looming on the horizon. Nvidia is having a hard time selling its core logic chipsets into either AMD or Intel CPU-based designs, with yesterday’s Apple news being a rare good-news exception for the company. And Nvidia seemingly still hasn’t found a customer for its Tegra ARM-plus-multimedia SoC for handhelds. Yet, with the exception of one notable layoff last September, the company has maintained its large and well-compensated workforce. What are all those engineers slaving away on?

I’ve long suspected that Nvidia’s been brewing an x86 CPU, as have others, and yesterday the company addressed the suspicions in a roundabout way, courtesy of senior VP of investor relations and communications Michael Hara’s comments at the Morgan Stanley Technology Conference in San Francisco:

I think some time down the road it makes sense to take [editor note: to the general purpose microprocessor business] the same level of integration that we’ve done with Tegra…Tegra is by any definition a complete computer on a chip, and the requirements of that market are such that you have to be very low power, very small, but highly efficient. So in that particular state it made a lot of sense to take that approach, and someday it’s going to make sense to take the same approach in the x86 market as well…If you look at the high-end of the PC market I think it’s going to stay fairly discrete, because that seems to be the best of all worlds…A highly integrated system-on-chip is going to make sense in the MID [editor note: mobile intelligent device] and netbook markets…Two or three years down the road I think it’s going to make sense…We won’t talk much more about what we think about that timeframe, but there’s no question it’s on our minds.

Do the math: for Nvidia to enter the market with an x86-based Tegra-like SoC (neatly validating Intel’s Atom aspirations for handhelds in the process) in the next two to three years likely means that development is already underway. I can’t help but be struck with the irony of this news, given always-flamboyant Nvidia CEO Jen-Hsun Huang’s comments less than a month back:

The heart of this issue is that the CPU has run its course and the soul of the PC is shifting quickly to the GPU.

If that’s true, Jen-Hsun, then why bother developing an x86 CPU? The answer, of course, is that Jen-Hsun knows his comments are little more than BS. Granted, there’s little indication that graphics processing will migrate back to CPU, in spite of a seeming plateau in mainstream applications’ polygon and pixel demands. GPUs also fill an important role in handling video processing in a speedy and power-thrifty manner. And there’s some possibility, though by no means a guarantee, that over time GPUs will take over additional functions amenable to their massively parallel processing nature, by virtue of coming-soon APIs such as OpenCL and DirectX 11. But for anyone who thinks that the GPU will obsolete the CPU or even tangibly retard its advancements in the future, I have seven words for you:

I’d like some of whatever you’re smoking.

Assuming my between-the-lines reading of Hara’s teasing is on track, where will Nvidia’s x86 capabilities come from? While I was pulling an April Fool’s joke when I ‘announced’ last year that Nvidia was buying AMD, the idea of Nvidia acquiring Via has been bantered about in the press for quite some time. There’s a problem with that prognosis, however, and it’s a biggie: Via’s x86 license isn’t capable of being adopted by an acquiring entity. Nvidia could try to do a straight-up x86 design, but (barring FTC pressure, which is perhaps what Nvidia’s gambling on), I can’t imagine a scenario in which Intel would grant its competitor a license. Emulation is also a possibility, but the track record here is spotty at best; Montalvo Systems flamed out before it ever produced its first chip, and Transmeta did little better, at least in the marketplace (in the courts, on the other hand, it fared somewhat better).

This feels like a ’swing for the fences’ (or, if you prefer, ‘Hail Mary pass‘) bet-the-company move on Nvidia’s part. And any of you who follow sports already have a sense of how often (or more accurately, ‘rarely’) home run and throw-for-the-end-zone attempts succeed, no matter how exciting they may be for us to watch. Your thoughts, readers?