We're here at CES 2017, and watched Jen-Hsun Huang's keynote address at the opening of the show - here's what he just announced.
Geforce Experience Connects to Facebook Live.
Since the Geforce platform is already connected to streaming and other social networks, and now there's a two-click option to connect directly to Facebook. And to show it off, Huang in full showman mode invites Bioware's Aaryn Flynn to the stage - because why not show this new feature with Mass Effect: Andromeda?
So what we were shown was real footage, captured on a GTX 1080, streaming direct to Facebook Live - and it was some pretty impressive new gameplay, to boot. New worlds, sexy space-ships, stirring music... All pretty odd for a CES keynote, which have never really focused so much on gaming before - but it certainly got a good crowd reaction!
It was all very smooth and seamless, but from a demo it's hard to tell how smooth it really will be - and it effectively makes Facebook Live a competitor to Twitch. More ways to share games is always good, but with Twitch already dominating the market, it will be interesting to see how this takes off.
This one's aimed at gamers who may be using older hardware without upgrading. Nvidia's been working on this for years, showing it off at E3 after E3 - in fact, most have thought it's little more than vapourware, but... In the words of Huang, "We've finally done it."
GeForce Now turns any PC into "the most powerful gaming PC, and it's all in the cloud." It's apparently a small download, too.
The demo showed the software off not only on a PC, but also on a Mac, with a game that's known for not playing well with integrated graphics - Tomb Raider. And it loaded smoothly, and looked great.
GeForce Now is coming in March, on a per-play basis. It costs $20 for 20 hours of play, and will have a range of performance levels. So, while it sounds like it might be a competitor to Nvidia's own hardware, it really is a whole other usage model. If you play a LOT of games, it's just not worthwhile, but if you're more casual, then it's a really solid option to get access to a lot great games.
And on to.. Shield!
The first big news - apart from the new hardware and form factor, is that 4K Netflix, Amazon, and YouTube connectivity is coming. On top of that, the device can stream over an Android Steam app 4K gaming to your PC.
Huang also talked up Amazon Echo, which is a little less exciting for those in regions with less than stellar Amazon access. But what this does is turn your television, via the Shield unit, into a home automation and AI agent, with Google Assistant.
"Wouldn't it be nice, if your AI was completely ambient," Huang asked, alluding to being able to speak to your Assistant anywhere in the house. The Nvidia Spot plugs into your wall, with all the computing happening on your Shield - so you can spread out the small microphone units everywhere.
"If you have multiple devices in a large room, it can even triangulate where you are biased on beam-forming," Huang said. And it can work out who's talking in a crowded room.
A video demo showed someone's daily routine - making coffee, checking the news, selecting a TV show to watch - but all powered by Shield and Nvidia Spot units. "In the future," Huang said, "Our house is going to become an AI. Iron Man's Jarvis is coming to us all - I decided we should build it for all of you."
And the big surprise, is the Shield's price. - $US199. No pricing yet for the Spot.
Huang kicked off talking about the gross costs and waste inherent in our current transportation systems, especially when it comes to moving consumables about, and the wastage of parked cars. He then segued into self-driving cars and GPU deep learning, but added that driving cars seems easy, but is hard to teach ta computer...
"Deep learning," he said, "has made it possible to crack that nut. We can now perceive our environment surrounding our car; we can predict using artificial intelligence where other people and cars will be."
"We can even teach a car how to drive, by watching and learning from us. And," Huang added, "the cloud can help us compare how we're learning."
Getting this all working together is a problem, but Nvidia's new solution is called Xavier - a computer about the size of two Raspbery Pis. "This runs a new operating system, called Driveworks.."
The unit contains eight ARM64 cores, and boasts performance of 30 trillion operations per second. The video demo of some rather nervous-looking chap driving to Starbucks nearby, and driving on a freeway no-less, suggests Xavier hooks up to six cameras around the car, and the hardware definitely has six outputs.
"These AI connected cars should be able to drive from address to address in nearly every part of the world," Huang said. But it can also work out when it has low confidence in its ability, and then hand off driving to the human - this is called the AI Co-pilot.
"It's aware of its surroundings, of the state of the driver, and of the passengers," Huang said. Combined with speakers inside the car, "The AI is either driving you, or looking out for you. It is still completely involved in the process."
Interestingly, the next demo showed Xavier running on only four cameras, but also showed off Xavier warning the driver of a number of possible hazards - that the driver may not notice. Merging traffic, approaching pedestrians or motorcyclists. Xavier can recognise faces, track your head, read your lips, and even track the focus of your gaze - all via deep learning.
All up - and I say this as a confessed non-driver - the Nvidia AI car platform looks pretty comprehensive. Huang also predicts the capability for the system to converse with home systems.
Nvidia's working with a number of partners on Xavier: Chinese mapping company Baidu, to map the world's largest market; TomTom, to handle the same thing in Europe, and Zenrin, the leading Japanese mapping company. Huang Also announced a partnership with Here, for the US market, and with ZF, a leading auto-supplier, ahead of a commercial solution coming to market in 2017.
The big announcement was Bosch's adoption of Nvidia's computing platform. Bosch services every car company in the world, so that's no small thing.
Oh, and Nvidia's working with a small company called Audi to "Build the next generation of AI car." No biggy. "We'll have cars on the road by 2020."
And, sadly, that was it - no news of the rumoured GTX 1080 Ti, though it could well still be announced post-show.