At a small press gathering at Computex early Monday morning, Nvidia CEO Jen-Hsun Huang bounded into the room, eventually eschewed his microphone, and showed off each of his company's most recent major product launches (the GeForce 980 Ti, Nvidia Shield Android TV Console, and Drive PX) like a proud papa boasting about his offspring.
Huang added that Nvidia is out of the smartphone business. All the better to focus more on gaming and some of the other industries the company's visual computing can foster, such as its traditional enterprise business and in automobiles.
Although his comments, both prepared and in response to a Q&A, didn't reveal any grand product news -- or at least none that Tom's Hardware hadn't already covered -- there were a few helpful clarifications around Nvidia's technology roadmap, including why the company is abandoning the smartphone business while still heavily embracing Android, along with information about its unique view of the Internet of Things (IoT), the cloud, and finally the future GPUs.
Nvidia Moves Out Of Smartphones, Into The TV
Instead, Nvidia has switched its focus to the living room and to Android, which Huang said he thinks will become the next big gaming platform, and in fact the future of gaming. He also characterized Android not as a mobile OS, but an Internet OS, and said that its openness and broad distribution were some of the reasons Nvidia chose it for its future gaming platforms.
The market for Nvidia's Shield family is massive, Huang said, adding that he would be disappointed if the company didn't sell tens or hundreds of millions of units over time. If that sounds effusive, Huang also tempered that statement by admitting that Nvidia's efforts were merely beginning, and that the company has plenty to learn about the manufacturing and distribution aspects of products like these.
But make no mistake, Nvidia is committed to developing an ecosystem around Shield. Huang brazenly called the Shield Console the future of television.
Maxwell, Pascal, And Beyond
Titan X, with its 12 GB of memory, is important for 4K video editing, for deep learning, and as a platform for researchers and developers (including those who are building 4K games). In other words, it's for those who need its memory.
Because Maxwell was Nvidia's first architecture designed for mobile first, but with the ability to scale up to massive parallelism in GPUs like the GM200, I asked Huang if he saw that same approach as the key to future architectures. The answer was a definitive yes. He called the decision to build Maxwell this way "one of the better strategic decisions we've ever made"
Even with thousands of people working on a next-generation graphics platform, it's expensive to build distinct architectures for the PC, cloud and mobile, he said. Building one architecture that is software-compatible across everything not only simplifies the design, but also all of the complex software (instruction set, middleware, compilers, and so on), and it therefore increases the velocity of execution.
Huang also indicated that building it for both the lower power demands of mobile and the high performance needs of desktop (and even cloud applications) creates what he called a fantastic dichotomy for the engineers, an interesting and useful tension that served Maxwell nicely. The energy efficiency of Shield helped drive the performance of the GeForce GTX 980 Ti, he said, and vice versa.
We wrote about the next generation graphics architecture, Pascal, based on Huang's keynote address during Nvidia's GTC. He said nothing new, but we pressed him on the next process node, and technologies such as high bandwidth memory. On the first point, Huang said that if he were a betting man, Pascal, and not iterations of Maxwell, would be the target for a new process node like 16 nm.
He said the overall strategy would follow a ping pong pattern. (Like tick tock?) The company is likely to create a new CPU architecture every other year, a new process technology every other year, with the same cycle for GPUs as well.
When asked about high bandwidth, or stacked memory for Pascal, Huang quickly touted the Maxwell memory architecture and then said that what matters is performance and efficiency. The future will likely hold stacked memories, but for now the cost of that memory is high and the availability is low. "It's a little too early to use it," he said.
Cloud, Auto, And IoT
Huang was asked about 4K gaming in the cloud, to which he responded with great enthusiasm, but he also pointed out that the infrastructure is still the complicating factor.
He said that cloud gaming's killer application wasn't an application at all, but the use case for cloud: the click and play convenience of it. No more disks, no more stores, no more long downloads. "Convenience," he said, "is the future of all computing." However, he predicted that we're still a couple of years away from having the technology, infrastructure and the end to end experience.
Nvidia has continued to invest in the cloud, not just for gaming but for anything that requires heavy parallel processing. The cloud, he said, is Nvidia's fastest growing business (60 to 70 percent annual revenue growth) and that it's now about a several hundred million dollar business.
While Nvidia is clearly no longer in the smartphone business, it is focused on auto -- probably one of the few areas, besides the enterprise business, that isn't gaming oriented. Huang described automobiles as visual super computers on wheels, and Tegra as that visual super computer.
Holding up the Drive PX board, Huang called it the world's largest IoT, but he also said that Nvidia's role in IoT would be on the larger side. That is, Nvidia will not be getting into the more miniaturized side of IoT we're accustomed to thinking of -- the Internet-enabled clothes and smart devices. These sensors and tiny SoCs will be sold by the trillions, he said, and that the market doesn't need that many companies. "We don't need to play in the ultra small segment," Huang added.