“We are thrilled by the growth of the ecosystem we are building together and will continue to put our heart and soul into advancing it. Building tools for the Da Vincis of our time is our purpose. And in doing so, we also help create the future.”
On the second day of the NVIDIA GTC 2021 event, CEO Jensen Huang presented the keynote speech, giving a presentation on its various projects. In the event that lasted close to two hours, Huang made announcements across sectors and domains which could transform many trillion-dollar industries. However, simulation, Omniverse, virtual avatar featured very prominently, staying true to the Metaverse rage currently in the tech space.
Some of the notable announcements are listed below.
Omniverse — A Great Example of Digital Twin
NVIDIA Omniverse is the company’s shot at creating an easily extensible and open platform for virtual collaboration and real-time and accurate simulation. Huang said in the keynote speech that with Omniverse, users will be able to create new 3D models of the physical world. With the newly announced Omniverse Avatar, developers will be able to create interactive characters that can see, understand, and interact on a wide range of topics.
Further, the new update to Project Maxine allows Omniverse Avatar to connect computer vision, Riva speech AI and animation to a real-time conversational AI robot. Project Maxine can also add state-of-the-art video and audio features to virtual collaborations and content creation applications.
Huang also introduced NVIDIA Omniverse Replicator, a synthetic data generation engine to train deep neural networks. There are two kinds of replicators — Omniverse Replicator for Issac Sim and Omniverse Replicator for DRIVE Sim for autonomous vehicles.
Full Stack and Open Platform
Huang spoke in length about Accelerated computing, which uses specialised hardware for exponentially speeding up the work by performing parallel processing of frequently occurring tasks. He said, “NVIDIA accelerated computing is a full-stack, data-centre-scale and open platform.”
In this regard, Huang spoke about three chips — GPU, CPU and DPU — and systems — DGX, EGX, RTX and AGX — spanning from cloud to the edge. NVIDIA also announced 65 new and updated SDKs at GTC.
Huang introduced NVIDIA Quantum-2, branding it as the most advanced networking platform ever built. He said that with BlueField-3 DPU, the company announces cloud-native supercomputing. Some of the features of Quantum-2 are high performance, broad accessibility, and strong security. Speaking of security, Huang also introduced a three-pillar zero trust framework to tackle cybersecurity challenges.
Over the last four years, the number of companies and partners in the NVIDIA Issac ecosystem has grown to 700, a five-times rise. NVIDIA has announced the Isaac robotics platform to further equip partner companies, which can be easily integrated into the Robot Operating System (ROS), a set of widely used software libraries and tools for robot applications.
For the NVIDIA Omniverse, Huang spoke about Isaac Sim, one of the most realistic robotics simulators available. Isaac Sim Replicator can help in generating synthetic data to train robots. This Replicator simulates the sensors and generates automatically labelled data. With a domain randomised engine, Replicator can create rich and diverse training data sets.
AI Models and Systems
The announcement of GPT-3 last year created major ripples in the AI world. Since then, many such language models have been introduced, each progressively larger and more advance than the previous. Now, Huang introduced Nemo Megatron to train large language models, which will reportedly be the biggest mainstream HPC application ever.
Huang also introduced NVIDIA Modulus to build and train machine learning models that obey the laws of physics.
Further, Huang introduced three new libraries — ReOpt, cuQuantum, and cuNumeric.
Huang announced NVIDIA’s efforts in the autonomous driving space. NVIDIA DRIVE, the company’s full-stack and open platform for autonomous vehicles, will now have Hyperion 8. NVIDIA’s latest complete hardware and software architecture suite includes 12 cameras, nine radars, 12 ultrasonics, and a front-facing LIDAR; two NVIDIA Orin SoCs process these sensors.
Hyperion also contains several new technologies built-in that includes Omniverse Replicator for DRIVE Sim. NVIDIA is currently running Hyperion 8 sensors, deep-learning-based multisensor fusion, feature tracking, and a new planning engine.