Nvidia Computex Press Conference, Huang Announces Jetson Xavier


We're here at Nvidia's Computex Press Conference here in Taipei. We'll update this article as Nvidia CEO Jensen Huang expands on the company's announcements, so be sure to refresh your browser, or periodically revisit this article. 

We're hoping Huang announces the company's next-gen graphics card, which is rumored to be named either the GTX 2080 or GTX 1180, but nothing is guaranteed.

1:00AM ET: Nvidia is taking its GTC tradeshow all over the world. Huang is fresh off his presentation at GTC Taiwan.

1:01AM ET: Jensen is providing an update on GeForce.

1:03AM ET: Jensen is displaying the difference between old gaming laptops and Max-Q models. Older models are 51mm thick and weigh 10 lbs., while new models are 17.7mm thick, weigh 4 lbs., and provide three times the performance of a PlayStation 4.

1:05AM ET: Nvidia has 26 million Max-Q laptops in the market. The company has 26 designs total in the pipeline, which is an increase over the eight models currently on the market. 

1:08AM ET: New Battle Royale-type games are expanding the gaming market. Battlefield V, Call of Duty Black Ops 4, and Shadow of the Tomb Raider are also coming to market.

1:10AM ET: Jensen says two new dynamics are changing the market: slower CPU performance increases, and the rise of Deep Learning and Artificial Intelligence. In the future, software will write software.

1:12AM ET: Applications will need a thousand times more performance in the future. GPUs are a natural fit.

1:13am ET: Jensen is outlining the DGX-2 supercomputer. "The World's Largest GPU." Two petaflops of performance in one node, 512GB frame buffer. Replaces 300 servers.

1:13AM ET: DXG-2 server has set five world records. Fastest single chip, fastest single node, fastest at scale, fastest inference, and highest inference throughput.

1:16AM ET: Jensen displays the HGX-2. A reference hyperscale design that allows OEMs/ODMs to build standardized supercomputers. Eight V100 GPUs in the SXM3 form factor, six NVswitches. One petaflop of performance.

1:25AM ET: Nvidia has enabled real-time raytracing. Jensen is displaying the Star Wars real time raytracing demo that we seen at CES.

1:30AM ET: Jensen is displaying real-time scene manipulation with raytracing.

1:30AM ET: All phones in the briefing room are buzzing, announcing Taiwan's air raid drill. Jensen pauses.

1:35AM ET: Three things have to happen to bring AI to the outside the world: AI needs great sensors to detect its environment, an AI brain that is high performance and lower power, and new actuators.

1:35AM ET: Xavier is key to enabling these goals. Jensen claims this is the most complex SoC ever made, 9 billion transistors, 350mm2, built on 12nmFFN.

1:36AM ET" Xavier powers Nvidia Drive Xavier (left) and Nvidia Drive Pegasus (right).

1:38AM ET: Nvidia is announcing the brand new Nvidia Isaac Robotics Platform.

1:40AM ET: Jetson Xavier. Eight cores, 9 billion transistors, 12nm FinFet. Single largest project in Nvidia's history. Five years ago they set specifications, four years to design, two years to architect. Designed to power robots.

1:41AM ET: The final package includes eight high-end arm cores. All in 30W. Essentially a the power of a 1000W workstation ($10,000) within a 30W envelope (per Nvidia). Early access now. To developers in August. Development kits are $1,299.

1:43AM ET: Nvidia Issac includes three components: The Jetson processor, MIX software, and virtual reality testing.

1:46AM ET: Nvidia's new "I Am AI" campaign, feels reminiscent of Intel's "AI On IA" slogan.

1:50AM ET: Jensen is recapping, opening the floor for questions. No next-gen GPU announcement yet.

1:55AM ET: Q. Is the processor in Jetson the same as Xavier? A. Yes, similar architectures for Jetson, Xavier, Pegasus, HGX-2, DGX-2. Insures  end-to-end architecture that eases programming.

1:50AM ET: Q. Which is better, PUBG or Fortnite? A. Jensen says both are great.

1:50AM ET: Q. Is GPU supply normalizing? A. Yes, supply has stabilized.

2:02AM ET: Q. How long until autonomous machines come to market, and what excites you the most? A. Autonomous vehicles are the first wave, will reduce accidents.

2:06AM ET: Q. When will real time ray tracing come to the mainstream? Still requires four Volta V100's to run now. A. Will take some time to shrink the compute power down to a smaller device. Game developers and film editors can use this now to speed image production, eventually it will come to the mass market.

2:10AM ET: Q. Hint when we might be able to run 4K 120Hz on a single card? A. Jensen could predict it, but he won't.

2:11AM ET: Q. When is the next GeForce coming to market? A. A long time from now. Jensen promises he will invite us out for the launch.

2:13AM ET: Q. The industry is moving towards heterogeneous CPU+GPU processing. Where does that leave Nvidia? A. Huang says that the best ratio of CPU and GPU compute changes based on the workload, so disaggregating the two resources is best.

This thread is closed for comments
    Your comment
  • ElectrO_90
    Is it me, or am I the only one underwhelmed with the comparison of an "old laptop" and a "new laptop".
    Couldn't we say exactly the same thing about mobile phones as well as many other pieces of equipment, ever made....
  • pwninsarge
    I think he answered a long time from now for the next gen gpus to dissuade more questions like that from being asked.
  • bit_user
    1920539 said:
    Max-Q models ... provide three times the performance of a PlayStation 4

    up to 3x the performance of the original PS4 (not PS4 Pro). Not all Max-Q models are that fast.