Nvidia CEO talks down CPU-GPU hybrids, Larrabee


— 12:28 PM on April 11, 2008

Yesterday at its Financial Analyst Day conference, officials from Nvidia talked for quite a while (about six hours) about the firm's position in the market. They addressed, at some length, how Nvidia plans to counter the AMD and Intel CPU-GPU hybrids and Intel's upcoming Larrabee graphics processor. Nvidia CEO Jen-Hsun Huang was quite vocal on those fronts, arguing hybrid chips that mix microprocessor and graphics processor cores will be no different from systems that include Intel or AMD integrated graphics today.

According to Huang, development cycles for processors are longer, and customers will continue to find value in pairing integrated or discrete Nvidia graphics with a processor that already includes a graphics core. Besides, Huang believes Intel's promise of ten-times-greater integrated graphics performance by 2010 will yield hardware barely at the level of current mainstream Nvidia GPUs. Nvidia's vision is that consumers will need relatively powerful GPUs as raw processing devices in the future, too. Later in the conference, software firm Elemental Technologies gave a demonstration of its H.264 video transcoding software—the kind you need to convert videos for an iPod—running about 19 times faster on an Nvidia GPU than on a quad-core processor. Huang thinks video transcoding in such instances ought to be "instantaneous" in the future.

Huang showed a slide suggesting consumers spent 7% less on processors and 14% more on GPUs last year compared to 2005. Nvidia believes users will get a better experience from pairing a less powerful CPU with a more powerful GPU, and Huang cited the commercial success of Gateway's P-6831 FX—a $1,349 laptop that pairs a 1.66GHz Core 2 Duo with a GeForce 8800M GTS—as evidence. The laptop has been hailed as the "best midrange gaming notebook ever" by Anandtech, and it reportedly sold out well ahead of schedule at Best Buy. The success of GPUs will only increase in the future, Huang projected:

Of course, all this talk about powerful graphics processors capturing more of the market wouldn't be complete without a mention of Larrabee, Intel's upcoming discrete GPU.

Huang summed up Nvidia's position on Larrabee in one sentence: "We're gonna open a can of whoop-ass [on Intel]." The Nvidia CEO's other arguments weren't quite as strongly worded, but he explained that Nvidia is continuously reinventing itself and that it will be two architectural refreshes beyond the current generation of chips before Larrabee launches. Huang also raised the prospect of application and API-level compatibility problems with Larrabee. Intel has said Larrabee will support the DirectX 10 and OpenGL application programming interfaces just like current AMD and Nvidia GPUs, but Huang seemed dubious Intel could deliver on that front.

Huang even argued that Larrabee's x86-derived architecture is a rehash of Itanium—the ill-fated "successor" to x86 processors—in the way it throws away "billions of dollars of investment" on current graphics architectures. Huang said CUDA, Nvidia's C-like application programming interface for general-purpose GPU computing, is already very successful, and suggested Intel might have a hard time displacing it with a completely different product based on a radically different architecture. Nvidia VP Tony Tamasi chipped in: "Larrabee is a PowerPoint slide, and every PowerPoint slide is perfect."

Tamasi went on to shoot down Intel's emphasis on ray tracing, which the chipmaker has called "the future for games." He prefaced his criticism by mentioning that the popular mental ray ray tracing renderer is owned by Nvidia, and that he has nothing against ray tracing in particular. However, his view is that ray tracing is just another tool in the toolbox and that the future of real-time graphics will involve a mix of rasterization and ray tracing—the very same approach used in movie rendering by the likes of Pixar. Additionally, Tamasi believes rasterization is inherently more scalable than ray tracing. He said running a ray tracer on a cell phone is "hard to conceive."

So, with Intel about to jump in the discrete GPU market, will Nvidia counter by moving into the microprocessor market? Huang hinted such a move isn't part of Nvidia's strategy for the time being. "We're gonna be highly focused on bringing a great experience to people who care about it," he explained, adding that Nvidia hardware simply isn't for everyone. He illustrated that point with two examples. One was Google, which he said doesn't need to make an operating system to compete with Microsoft. Another was the iPhone, which Huang thinks isn't as good as a Blackberry for e-mail, even though it does have e-mail functionality. Nonetheless, Huang added, "I would build CPUs if I could change the world [in doing so]."

Even if Nvidia doesn't care to make CPUs, it doesn't mind taking microprocessor market share away from Intel—either directly or indirectly. One crucial battlefield in the coming months and years is the low-cost and mobile stage, and Huang argued Nvidia is armed and ready on that front. First, Nvidia is readying a platform to accompany VIA's next-generation Isaiah processor, which should fight it out with Intel's Atom in the low-cost notebook and desktop arena:

Another of Nvidia's weapons is APX 2500, the applications processor for mobile devices scheduled to hit production later this quarter. The APX 2500 squeezes an ARM11 microprocessor core, a high-definition audio/video processor, a GeForce graphics processor, and a DDR memory interface into a package smaller than a dime, with power consumption low enough to allow for up to ten hours of continuous 720p high-definition video playback. The APX 2500 (or a future iteration of it) may compete with the system-on-a-chip successor to Intel's Atom, which is code-named Moorestown and is scheduled to appear in 2009 or 2010.

   
Register
Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.