Nvidia presented an intriguing vision of a low-power computing future at its GPU Technology Conference in San Jose this week, where it announced “Kayla,” billed as “the world’s most powerful ARM computer.” Nvidia also unveiled the first Tegra processor that will support its CUDA parallel-processing technology, as well as the first chip to support “Project Denver.”
Nvidia has long been known as a component manufacturer, having helped pioneer the 3D graphics chip alongside such defunct manufacturers as Rendition and 3Dlabs. Over time, Nvidia has broadened its offerings to include both mobile chips as well as products for workstations and servers. In recent months, it has even become a full-fledged systems manufacturer, unveiling its first VCA appliance for “remote graphics” as well as an updated enterprise GPU roadmap that includes two new chips, “Maxwell” and “Volta.”
While the company’s latest enterprise cores offer the promise of high margins, sales into the high-volume mobile market—such as phones, tablets, and other intelligent devices—may be Nvidia’s future.
“We felt that the world was changing, that we were going to put computers in everything,” company’s chief executive, Jen-Hsun Huang, told the audience at the conference. “The idea is that it’s now in servers, and in PCs, but in the future computers were going to disappear, disappear that in the sense it would be everywhere, literally everywhere.”
Given that future, he added, “we wanted to invest in processors that had the same visual beauty, the same interactive capability that we’re known for but in computers that are literally everywhere: in stores, in glasses in watches…anything that has a display in the future will have a visual computer inside.”
Nvidia’s first Tegra chip, by Huang’s own admission, didn’t turn out that well. The company improved on Tegras 2 and 3, with the latter employing a dedicated low-power core. The Tegra 4 added computational photography, allowing objects to be tracked by a smartphone camera.
Nvidia’s next Tegra is code-named “Logan,” which includes a “Kepler” GPU, with full CUDA 5 support, plus OpenGL 4.3 support. “It does everything a modern computer ought to do,” Huang said, adding that Logan will be in production in early 2014.
The next Tegra chip, “Parker,” brings three ideas to the market, Huang said. It’s the first to include “Denver,” a combination of an ARM processor and a next-generation GPU—in this case, Nvidia’s Maxwell chip.
“To enable phones, tablets, and low-end servers, Parker will include FinFET transistors. “What’s really remarkable here is that, in five years’ time, we’ll have increased the performance of Tegra by 100 times,” Huang said. “Moore’s Law would suggest about eight. That’s a perfect example of disruptive technology… You should expect that in every generation, the Tegra processor brings something that is enormously surprising from the past.”
But Nvidia wanted to enable ARM computing as soon as it could. “So the guys went and said, we want to see CUDA in ARM right now, so we did,” Huang added. “We went out and built a new GPU, a GPU that you guys have never seen, and we’re going to combine it with ARM to offer the world’s most powerful ARM computer.”
And what was it? “Kayla,” a board combining Nvidia’s Tegra chips, which will merge both Logan and a Tegra 3 with PCI Express. Huang said that Logan will be the size of a dime and require no fans or heat sinks; he demonstrated the system’s ability to perform real-time ray tracing, just like its desktop GeForce chips.
Will Kayla enable a new breed of GPU-powered, low-power computers? Time will tell. Nvidia didn’t give a release date for Kayla—but if it is anywhere near Logan’s release date in 2014, things could get interesting.