Nvidia’s Jen-Hsun Huang on the adoption of VR, self-driving cars, and deep learning

The PC industry is slowing, but graphics chip maker Nvidia blew past Wall Street estimates for its second fiscal quarter. Jen-Hsun Huang, the colorful chief executive of Santa Clara, Calif.-based Nvidia, said in an interview with GamesBeat that the performance was helped by the smooth launch of the company’s next-generation graphics architecture, dubbed Pascal.

Nvidia launched four Pascal-based graphics processing units (GPUs) during the quarter, with a product slate that included a 15-billion-transistor deep learning chip. And Nvidia skirted the PC industry’s woes by focusing on more targeted markets, including gaming, professional visualization, data centers, and automotive electronics.

[aditude-amp id="flyingcarpet" targeting='{"env":"staging","page_type":"article","post_id":2028080,"post_type":"story","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"business,games,pc-gaming,","session":"D"}']

Nvidia talked about these targets during its earnings call, and I followed up in a conversation with Huang on Thursday after the call. Here’s an edited transcript of our interview.

GamesBeat: When you were saying in the analyst call that Pascal was as good a launch as you’ve ever done, what did you mean by that, in terms of some of the details?

Jen-Hsun Huang: We ramped into production four GPUs, each one of them more perfect. Four GPUs in one quarter, four separate processors in 16nm FinFET is quite unprecedented. It also uses the world’s fastest memory: HBM2 in the case of GP100 to G5X for Titan and 1080. We also use 3D-stacked memory. All of this happened together on Pascal. We did an extraordinary job. The functionality is also brand new. It’s the world’s first graphics pipeline that has multi-projection. It also has brand new deep learning architecture. All of these things combine, along with the organization’s execution in one quarter. I’m incredibly proud.

GamesBeat: How is VR adoption looking?

Huang: Our connection rate to VR displays, as you can imagine, is very high. We track it. We know very well where they are. The rate is growing very nicely. I can tell you that we’re quite delighted.

One of the great surprises I’ve seen is the success of Nvidia’s Fun House. It’s been downloaded so much. It’s quite something. People are obviously enjoying it.

GamesBeat: For Pascal, do you think hardcore gamers are driving it more, or VR people? Or are they the same people?

Huang: It’s mostly just gamers. The thing I mentioned, and you know this very well—the production value of video games has gone up so much. It’s the reason why your GPU adoption at the high end has increased. The next generation Scorpio and Neo are coming out. I’m super excited about that. Production value keeps going up.

[aditude-amp id="medium1" targeting='{"env":"staging","page_type":"article","post_id":2028080,"post_type":"story","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"business,games,pc-gaming,","session":"D"}']

GamesBeat: You’ve commented before on how strong the VR startup community is, as well as A.I. and deep learning startups. Are those as strong as you’ve seen so far?

Huang: The rate remains the same. The quantity continues. We’re seeing a lot of A.I. companies now in life sciences and health care…obviously, in transportation. We see a lot in fintech now, and some in enterprise software. The reach of A.I. is going to expand. In the area of VR, what’s exciting is we’re seeing a lot of great startups in haptics. As you know, haptics is going to be one of the most important areas of VR work.

GamesBeat: I noticed Intel bought Nervana. I wonder how you feel about the competition in deep learning against Intel.

Huang: Our focus is on what we should do ourselves. We have a good sense of the pulse of the industry. We know what’s going on out there. We know that deep learning adoption is broad. It’s going into production at scale. That’s happening right now. We know that this architecture, this new computing model, is much better with a very parallel architecture. Traditional computing architectures need to be augmented for deep learning We recognize that it’s a new computing approach and processes like GPU-to-GPU computing are a good option for it.

[aditude-amp id="medium2" targeting='{"env":"staging","page_type":"article","post_id":2028080,"post_type":"story","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"business,games,pc-gaming,","session":"D"}']

Our approach, as you know, is to take the GPU architecture we have and evolve it as quickly as we can to enhance the future of deep learning. We’re leveraging our scale to build these processors as fast as we can.

GamesBeat: How fast do you think the self-driving car phenomenon is going?

Huang: Autopilot is going to improve dramatically in the next year. I’m looking forward to real big advances in autopilot capability. I’m also expecting companies all over the world starting to deploy mapping and experimental cars for taxi services. You’ll see a lot more activity around that. You’ll also see a lot of trucks that will become more autonomous on highways. We’ll see big progress in the coming year or two, no later than that.