At its virtual GTC event, NVIDIA announced a number of new technologies it believes will transform several industries.
It expanded its virtual world collaboration platform Omniverse with the introduction of NVIDIA Omniverse Avatar and NVIDIA Omniverse Replicator.
NVIDIA Omniverse Avatar can be used to create interactive AI avatars using the company’s speech AI, computer vision, natural language understanding, recommendation engines, and simulation technologies. NVIDIA believes these avatars will be useful in customer service interactions such as restaurant orders, banking transactions, and making personal appointments.
NVIDIA Omniverse Replicator is a data generation engine that creates synthetic data that can be used to train deep neural networks. It is launching with two replicators, one of which is for NVIDIA DRIVE Sim and the other which is for NVIDIA Isaac Sim. According to the company, these two replicators will enable developers to bootstrap AI models and fill in data gaps.
“Omniverse Replicator allows us to create diverse, massive, accurate datasets to build high-quality, high-performing and safe datasets, which is essential for AI,” said Rev Lebaredian, vice president of simulation technology and Omniverse engineering at NVIDIA. “While we have built two domain-specific data-generation engines ourselves, we can imagine many companies building their own with Omniverse Replicator.”
Beyond the Omniverse, the company announced a number of other AI innovations at the event:
- Nemo Megatron, which can be used to train large language models.
- Deep Graph Libraries, which is a Python package for projecting graphs into deep neural network frameworks
- NVIDIA Modulus, which builds and trains physics-informed machine learning models
- An update to Triton, which is an inference server, that now can inference forest models and perform multi-GPU, multi-Node inference
Three new libraries were also introduced, including ReOpt, which is aimed at the logistics industry; cuQuantum, which will help accelerate quantum computing research; and cuNumeric, which accelerates NumPy for Python data scientists.
To further autonomous vehicles, the company announced Hyperion 8, which is a hardware and software architecture for autonomous vehicles. NVIDIA’s CEO and founder Jensen Huang said during his keynote, “By 2024, the vast majority of new EVs will have substantial AV capability.” Hyperion 8’s sensor suite includes 12 cameras, nine radars, 12 ultrasonics, and a front-facing lidar.
A full list of announcements from the event is available here.
