As a developer of graphics technology, Nvidia has been incredibly successful. Despite severe competitive pressure from AMD, Nvidia’s desktop GPUs still hold the number one market share, though AMD recently upped the ante with the release of the Radeon HD 5870 , which is hands down the fastest single GPU card today.
It’s clear, though, after Nvidia’s recent GPU Technology Conference, that the company’s aspirations lie well beyond building graphics chips. That’s not a revelation – Nvidia’s been saying this for several years now. For an industry observer, though, the GPU Tech Conference lays out Nvidia’s model for moving beyond just graphics.
That model, ironically, is Intel.
You won’t see Nvidia rushing out to buy or build new fabs, nor is it likely the company will bring to market an x86 CPU. Rather, what Nvidia’s trying to do is present the face of a modern, mature company with its fingers in many pies and its eyes firmly on future growth.
image credit: venturebeat
The press conference last Wednesday proved illuminating. It wasn’t so long ago that CEO Jen-Hsun Huang was throwing out phrases like “opening up a can of whoop-ass” when referring to Intel. By contrast, the press conference at the Fairmont Hotel in San Jose was positively restrained. Huang was still the centerpiece of the press conference, taking almost all the questions. But every time he’d start to stray into “whoop ass” verbal territory, he’d pull back just a bit. There was no doubt in anyone’s mind that he’s still very competitive. It’s also clear he’s become a little more cognizant of what Nvidia needs to do to stick around for the long term.
“We’re not a startup anymore,” he said at one point. “We’re a big company now.”
Everyone knows that, of course, but Nvidia’s always behaved like a startup. Now, it’s starting to think of itself as a large, mature entity. There’s a tendency to think of “mature” as some kind of pejorative in the technology business, but technology is mainstream now. Investors and analysts feel much more comfortable when a company acts like a grown-up.
The attitude extended to my conversations with other people at the conference. Take, for example, how the rollout of Nvidia’s new GPU technology, dubbed “Fermi,” was handled.
In a meeting with Tony Tamasi, Nvidia’s senior VP of technology and content, he acknowledged that they needed to say something to counter AMD’s recent Cypress (HD 5870 & 5850) announcements. But his tone when talking about Fermi was low key. When pressed about CUDA, Nvidia’s proprietary interface for writing general purpose GPU applications, he took the position that Nvidia was in no way positioning CUDA against emerging standards like OpenCL or DirectCompute 11.
“It’s about getting apps on the GPU, any way possible. CUDA, OpenCL, DirectCompute – they’re all good.”
Nvidia’s cool new add-in for Microsoft Visual Studio, Nexus, reinforces the point. Nexus works with CUDA, but also OpenCL and DirectCompute, allowing Windows developers to work natively inside Visual Studio, and picking their interface of choice to the GPU – all working seamlessly with CPU code.
Nvidia has lost none of its ambitions, though. Sanford Russel, the GM of the compute side at Nvidia, said it bluntly: “We’re a processor company, not a graphics company.”
But rather than blindly take on processor companies head on, as it seemed to want to do a couple of years ago, it’s taking a longer term approach, working with key software partners like Microsoft and Apple to gradually position the GPU as an equal to the CPU. That’s a much more indirect approach than trying to take on Intel directly.
Still, the road will not be easy. AMD has already shown it can build a great graphics processor with Cypress. As it moves slowly towards integrating CPU and GPU functionality, it will more directly compete with Nvidia on the GPU compute front – something it hasn’t placed a strong emphasis on in the past. Similarly, Intel is slowly – too slowly for some observers – building Larrabee, their take on a completely software programmable GPU.
Sanford Russell believes that Larrabee is too unbalanced toward the x86 side, and won’t be as good at the highly data parallel applications Nvidia is targeting with Fermi. In reality, no one really knows how good Larrabee will be. However, Intel has the will and the resources to pursue Larrabee until they get it right.
Right now, Nivdia is seeing some success with its much smaller Tegra line, and is starting to rack up some design wins. However, its chipset business is slowly fading away, since Intel has been unwilling to issue bus licenses for the Nehalem generation of CPUs. There’s Ion, of course, but that’s really a long term dead end, as Intel’s Atom transitions to become more of a system-on-chip.
So it won’t be an easy path for the other Santa Clara chip company. It is, however, a path that Nvidia needs to pursue to remain relevant and keep growing. The pressures, however, will be enormous.