“For 25 years, computer graphics has been the driving force of the company,” Huang said. However, there are other areas of innovation Nvidia is now driving, namely AI and robotics.
As Huang pointed out, most studios are still doing software raytracing via large, expensive renderfarms. Indeed, raytracing can, and is, being done using software, but the goal of Nvidia is to accelerate that domain of application.
With Nvidia’s Turing GPU architecture and its RTX real-time raytracing technology, it is no longer just about hardware vs. software rendering, but rather accelerated raytracing and accelerated rendering – real-time raytracing and real-time rendering. This enables artists to see realistic, high-resolution representations of their work while they are working, as opposed to viewing a lower-res representation while working, and then waiting hours for a high-res render from the renderfarm.
To this end, Huang listed more than 20 partners supporting RTX in their software, including Adobe, Autodesk, Dassault, and Pixar, as well as the two major game engine developers – Unity and Epic. Huang then provided a very informative as well as entertainment demonstration of the game Quake. Twenty years ago,
Quake ushered in the era of hardware acceleration. “Every company needs a killer app, and it was
Quake,” Huang stated. The demo was nostalgic. But then, the audience began to see the game in a whole new light, transformed with this latest technology.
As Huang noted, offline rendering is used in every industry today, from architecture, to product design. A modern movie, for instance, has 2,500 to 3,000 shots that are only a few seconds long, but it takes artists the entire filmmaking process to create them. “The datasets are so large. Finally, this year with RTX, we are accelerating their tools,” he added. In fact, a large percentage of toolmakers in studios have adopted RTX, and little wonder why: they can reap huge savings in time and cost.
There are more than 200 animation studios in the world today, Huang said, and they need to work together. They need a tool that will make it possible for them to collaborate, and that is Omniverse, which lets artists harness multiple applications to create and share scenes across different teams and different locations. He likened the tool to Google Docs for 3D graphics.
Omniverse is open platform and works with all the major 3D tools. So for artists working separately, one creating geometry in Maya and another texturing in Substance, for instance, they can see all the changes occurring in real time. And, Omniverse can run on a local workstation, in a data center, in the cloud….
To speed up the work of those creating high-end DCC, Nvidia has the RTX Server, an architecture featuring 40 Turing GPUs in 4U, 6U, and 8U configurations. Not only ideal for rendering with remote workstations, they are also geared for cloud gaming via GeForce NOW. “One billion people cannot play games as they were intended,” Huang said. With this, Nvidia is offering a GeForce PC in the cloud, basically.
Huang then announced the GeForce NOW Alliance, which expands the GeForce NOW online gaming service that is built around specialized pods packed with 1,280 GPUs in 10 racks, interconnected with Mellanox high-speed interconnect technology through partnerships with global telcom providers. The partners will scale GeForce NOW to serve millions of gamers.
Look out for high-quality gaming coming to 5G teleco networks in the very near future.
Chapter 2: AI and HPC
It is the fastest-growing field of computer science today, Huang said, and has become the fourth pillar of discovery. “Now, data science lets us solve problems that were previously impossible,” he said.
In fact, he noted that computers are now talking more to computers than to humans. And, that conversation model will continue to grow. “Data science is the new HPC challenge because the data is so large,” he added. That’s because of all the data analytics. To accelerate the data preparation, model training, and visualization, Huang introduced the data science workstation powered by Nvidia. It is configured with Quadro RTX GPUs and pre-installed with CUDA-X AI accelerated machine learning and deep learning software. And, they are built by global workstation providers.
What used to take scientists eight days to do can be done in minutes.
Chapter 3: Robotics
Huang announced a new robotics computer, the smallest one every built: the Jetson Nano. Priced at $99, is a CUDA-X AI computer and runs the CUDA-X AI stack. It delivers 472 Gflops of compute performance for running modern AI workloads. Unbelievably, it consumes as little as 5 watts of power but supports the same architecture and software running America’s fastest supercomputers.
It just wouldn’t be GTC without an update on autonomous vehicles. In this regard, Huang announced the DRIVE Constellation cloud-based driving simulation platform. He also announced Safety Force Field, designed to shield self-driving cars from collisions.
At the start and end of the keynote, Huang had the audience repeat the acronym PRADA, which stands for programmable acceleration of (multiple) domains in one architecture. He explained how this pertains to data science:
Make it programmable. By making the CUDA-X libraries software-defined, the computer is improving all the time, and in turn, the time to solution is shorter. The growth is within the domains – if it is programmable, it can be used for many things. And by using a software-accelerated approach, you can expand the domains you can use the data for.
Acceleration. It offers the lowest cost of infrastructure.
Domains. Build a whole ecosystem around it. The more domains it serves, the lower the cost of deployment.
Architecture. An application written yesterday has to run on hardware today. Backward compatibility expands the domain and drives the cost of infrastructure down, growing the architecture constantly.
Huang then summarized all the announcements for the 9,000 attendees. The keynote provided a break in the myriad sessions focused around the major topics and more covered in his presentation.