It’s better late than never! In March, NVIDIA hosted its largest GTC ever. Despite a personal loss that cut my visit short, I managed to experience the event and meet some of you. I regret missing my scheduled speech on Thursday and appreciate your understanding. The conference was a unique experience, and I’m grateful for the opportunity to be part of it. Thank you for your warm introductions!
Massive Time
Since quite a lot of that is “previous information” by now, I’ll level out a few of my observations from the present. First, the convention had a different feel than the earlier (2019) one. , the sheer variety of folks (estimated to be ~ 16.5K) put pressure on the assets of the San Jose Conference Heart. It felt like not less than twice the variety of folks because of the final one. The present has outgrown the venue. That is for several causes, the obvious being that NVIDIA is now valued at over a trillion {dollars}. This has reached a crucial mass th, where all monetary and know-how must concentrate.
There’s one other huge level. It’s the “present” half. Whereas Jensen Huang has more keynotes than you could shake a stick at, NVIDIA cranked it up a notch. NVIDIA rented out a sports activities environment for the keynote, with rock live performance-worthy stagecraft.
AI
The buzz around AI was palpable at the conference, permeating every conversation. The energy was electrifying, and it’s a testament to the power of tech marketing that there’s no such thing as a subtle AI discussion at these events.
Exhibit Flooring
The exhibit hall was a treasure trove of innovation, always bustling with activity. Numerous fascinating sites were to be explored, and the ‘theaters’ where lightning talks were held added an element of surprise. Whether it was a vendor or a researcher, each 10-minute talk offered a unique perspective on various topics.
The exhibit grounds were divided into totally different areas. The sting computing space was where many of the Jetson crowd were situated. There was additionally a robotics space where attention-grabbing work was being introduced. I had a nerdgasm once I noticed the Disney Analysis BD-X droids there:
I performed an in-depth analysis and guessed how they worked. Disney Imagineering held a session at GTC, which informed us precisely how they do! One attention-grabbing truth is that the droid’s gait uses a reinforcement-studying mannequin developed on ISAAC. It takes an RTX4090 ~48 hours to create the gait mannequin. There are two Jetson Orin NX within the droid. One within the head and one within the physique. This offers it sufficient com computing to handle very attention-grabbing issues! In any case, they’re analysis initiatives. Since GTC, these droids have been seen at Disneyland, working for a residing. Folks, particularly youngsters, love these items.
Unitree had some doggies there!
Classes
There have been hundreds of periods, so I can’t even start to tell you what was lined up. The beauty of periods is that you’ll find a subject that you’re curious about, curious about sessiointoabout it, and have a fantastic overview of the place the state-of-the-art is in that topic.
I’ll additionally point out that something that entails computation (nearly all science, know-how, and enterprise fields now) is represented. For example, I went to a session concerning the Sphere in Las Vegas, Nevada. They had been speaking numbers, and getting your head around is exhausting. For instance, there are ~ 166,000 audio systems within the venue. The audio system beams audio to any given seat. This implies each seat in the home has an optimum audio expertise. Getting the information to the right place in the community is an arduous process. The 50-minute present now enjoying is 14 petabytes (PETABYTES!) of uncompressed video and audio. They use NVIDIA switches to assist in transferring around. Undoubtedly, it’s enormous league stuff.
Bulletins
The conference was filled with exciting announcements, particularly those related to Jetson. These announcements are sure to impact the future of technology significantly.
Blackwell
Blackwell is the name of the brand-new structure for NVIDIA GPUs. Like earlier generations, NVIDIA will be providing these in many configurations. We’re still a bit far off from all of the technical particulars. Still, the information center GPUs will probably be two dies related by way of a high-speed transport material in the identical package, much like the structure of Apple M3 Professional chips.
As you recall, NVIDIA canceled the Atlan era of Jetson. This year, they introduced the Thor era. This will likely be a Blackwell structure, with 6-10X the efficiency of the Orin line. This can open up quite a lot of new utility areas.
Robotics
The primary Jetson Thor chips will probably be for automotive. In addition, they introduced Challenge GROOT, a general-purpose basis mannequin for humanoid robots. GROOT will run on Thor, with all the humanoid distributors (apart from TESLA) on board.
There’s a significant emphasis on digital twins. The idea behind digital twins is that you construct environments and robots in simulation that precisely replicate their actual-world counterparts. You practice the robots to work in that setting. Then, you deploy the educated fashions onto the true-world robots, and everything goes as planned.
There have been many real-world instances of working examples. Ultimately, this could save quite a lot of money and time.
Conclusion
GTC 2024 was pretty the expertise. If you need inspiration, consider attending certainly one of these occasions. Everyone seems to be there. There are lab periods where you may have any questions answered by the engineers who work on it day in and time out. Nicely definitely worth the funding. If there may be one tip I’ve, it is best to go over the periods rigorously and discover ones that you find attention-grabbing. Planning is essential right here. There are too many shiny and shiny to distract. However, these may be enjoyable, too!