As he does every year, Nvidia CEO Jen-Hsun Huang took the stage at the GPU Technology Conference in San Jose, California today, telling a packed audience how important the company's processors are for intensive computing tasks. He shook hands, received endorsements, demoed state-of-the-art computer simulations, and introduced a few new products.
Then, he went straight to the telephone, signed onto an investor call, and shared his endgame, the plan that ties it all together: Nvidia wants to take advantage of what he calls a "seismic change" in the computing industry to have its chips drive every single computer in existence. "Our strategy, ultimately, is to light every single pixel," said Huang. "There are 5.5 billion HD displays by 2017. We want to light every one of them."
"It's not just about phones."
Nvidia is a mobile processor company now, not just a GPU firm, and sure enough Huang hopes the company's Tegra chips will play a prominent role. That's putting it mildly, in fact: he told investors that there's a $10 billion market opportunity to address when counting the proliferation of tablets and LTE smartphones, with an estimated 500 million LTE devices on the market by 2017. But Nvidia's leader isn't just talking about putting chips in more mobile devices in addition to its existing desktop and mobile GPUs. "It's much more to us than mobile. It's not just about phones," he says, and the GPU announcements he made on stage today speak to that effect.
The next Tegra chip, Logan, due in 2014, isn't just more powerful than Tegra 4. It incorporates an actual laptop-class Kepler GPU, and comes with CUDA support. That allows developers to harness the GPU for actual computing, not just graphics performance, though there should be plenty more of that as well. It's more like a real computer than a smartphone, in other words.
And the Tegra after that, codenamed Parker and tentatively scheduled for 2015, will be all Nvidia's own design: a custom 64-bit "Project Denver" ARM CPU and a next-gen Maxwell GPU, built using a 3D transistor technology for even lower power, and performance estimated at 100 times the original Tegra 2. To show off what a next-gen Tegra could be capable of, Nvidia pulled out a "Kayla" developer board, which pairs an existing Tegra 3 chip with an unannounced Kepler mobile processor. He showed off real-time raytracing, smoke, and water simulations simultaneously on Kayla, and the message was clear: Nvidia isn't just about entertainment and mobility; it can build serious personal computers if it wants to, and it might be able to do so without help from Intel or AMD.
And yet, it's not clear that Nvidia wants to build computers at all, when it could potentially stream images and processing power from its server partners to any screen you might own. "If we don't power all of them, we can stream to the rest of them," Huang told investors. "We don't have to be inside the DVD player. We could be the server that's streaming DVDs to you on your television."
What he's talking about is Nvidia Grid, the company's play for cloud computing, which uses the company's desktop-grade GPUs with low-latency remote display technology to (hopefully) make it feel like a powerful distant computer you're using is in the very same room. Nvidia's first market will be gaming, but the enterprise is next: today, Nvidia announced the Grid Visual Computing Appliance, a server with eight Grid GPUs designed to provide the power of a workstation to any device for any employee in a company, splitting the power of several workstations across many employees, or giving any single person access to multiple workstations simultaneously.
Virtual workstations for everyone
He believes that there's a $5 billion Grid market opportunity to tap by selling such GeForce Grid hardware at nearly $25,000 each, or about $40,000 for a maxed rig, plus a yearly software license that should put upwards $300 in Nvidia's pocket per year for each GPU. To justify the price, Huang simply pointed to the massive confusion in the IT space about the proliferation of new platforms, and the the cost of outfitting employees with individual PCs. "You hire 20 consultants... they all have a computer on day one," he told investors. Dell, HP, IBM, Citrix, Microsoft, and VMWare are already working with Grid, building servers and thin client software. "75 large scale trials are happening right now as we speak," Huang says.
Here, too, Nvidia has the potential to push Intel and AMD out. Every company that buys into Grid is purchasing many Nvidia GPUs at the server end, but only a small number of rival CPUs to run the rig. Meanwhile, Grid users no longer necessarily need the power of Intel at their own personal computers since they can draw on Nvidia's cloud computing.
Though Huang expects Grid will be popular enough to contribute to the company's bottom line this year — as will Nvidia's Project Shield game console, he hopes — he cautions that the realization of this vision may still be a ways off. "We're not yet a cloud computing company. We're not yet a mobile computing company. We're not yet that company that we're trying to build."
"What we want to show you today is what's ultimately in our mind," said Nvidia's CEO.
In the meanwhile, Nvidia will keep doing what it does best: wow an audience with what its GPUs can do at technology conferences like this one. In a couple of years, Nvidia's Volta GPU will stack the memory on top of the processor itself for bandwidth of 1TB / sec, which Huang says is enough to process an entire Blu-ray movie in one-fiftieth of a second. No doubt it will be capable of even more impressive feats by GTC 2015.