Recent advances in large language models (LLMs) have fueled state-of-the-art performance for NLP applications such as virtual scribes in healthcare, interactive…
Recent advances in large language models (LLMs) have fueled state-of-the-art performance for NLP applications such as virtual scribes in healthcare, interactive virtual assistants, and many more.
To simplify access to LLMs, NVIDIA has announced two services: NeMo LLM for customizing and using LLMs, and BioNeMo, which expands scientific applications of LLMs for the pharmaceutical and biotechnology industries. NVIDIA NeMo Megatron, an end-to-end framework for training and deploying LLMs, is now available to developers around the world in open beta.
NeMo LLM service
The NVIDIA NeMo LLM service provides the fastest path to customize foundation LLMs and deploy them at scale leveraging the NVIDIA managed cloud API or through private and public clouds.
NVIDIA and community-built foundation models can be customized using prompt learning capabilities, which are compute-efficient techniques, embedding context in user queries to enable greater accuracy in specific use cases. These techniques require just a few hundred samples to achieve high accuracy. Now, the promise of LLMs serving several use cases with a single model is realized.
Developers can build applications ranging from text summarization, to paraphrasing, to story generation, and many others, for specific domains and use cases. Minimal compute and technical expertise are required.
The Megatron 530B model is one of the world’s largest LLMs, with 530 billion parameters based on the GPT-3 architecture. It will soon be available to developers through the early access program on the NVIDIA NeMo LLM service. Model checkpoints will soon be available through HuggingFace and NGC, or for use through the service, including:
The BioNeMo service, built on NeMo Megatron, is a unified cloud environment for AI-based drug discovery workflows. Chemists, biologists, and AI drug discovery researchers can generate novel therapeutics; understand their properties, structure, and function; and ultimately predict binding to a drug target.
Today, the BioNeMo service supports state-of-the-art transformer-based models for both chemistry and proteomics. Support for DNA-based workflows is coming soon. The ESM-1 architecture provides equivalent capabilities for proteins, and OpenFold is supported for ease of use and scaling of workflows for predictions of protein structures. The platform enables an end-to-end modular drug discovery workflow to accelerate research and better understand proteins, genes, and other molecules.
NVIDIA has announced new updates to NVIDIA NeMo Megatron, an end-to-end framework for training and deploying LLM up to trillions of parameters. NeMo Megatron is now available to developers in open beta, on several cloud platforms including Microsoft Azure, Amazon Web Services, and Oracle Cloud Infrastructure, as well as NVIDIA DGX SuperPODs and NVIDIA DGX Foundry.
NeMo Megatron is available as a containerized framework on NGC, offering an easy, effective, and cost-efficient path to build and deploy LLMs. It consists of an end-to-end workflow for automated distributed data processing; training large-scale customized GPT-3, T5, and multilingual T5 (mT5) models; and deploying models for inference at scale.
Its hyperparameter tool enables custom model development, automatically searching for the best hyperparameter configurations for both training and inference, on any given distributed GPU cluster configuration.
Large-scale models are made practical, delivering high training efficiency, using techniques such as tensor, data, pipeline parallelism, and sequence parallelism, alongside selective activation recomputing. It is also equipped with prompt learning techniques that enable customization for different datasets with minimal data, vastly improving performance and few-shot tasks.
Developers, creators, and enterprises around the world are using NVIDIA Omniverse to build virtual worlds and push the boundaries of the metaverse. Based on…
Developers, creators, and enterprises around the world are using NVIDIA Omniverse to build virtual worlds and push the boundaries of the metaverse. Based on Universal Scene Description (USD), an extensible, common language for virtual worlds, Omniverse is a scalable computing platform for full-design-fidelity 3D simulation workflows that developers across global industries are using to build out the 3D internet.
With these latest releases and capabilities, developers can build, extend, and connect 3D tools and platforms to the Omniverse ecosystem with greater ease than ever.
Cloud services for building and operating metaverse applications
The first NVIDIA SaaS offering, Omniverse Cloud, is an infrastructure-as-a-service that connects Omniverse applications running in the cloud, on-premises, or on edge devices. Users can create and collaborate on any device with the Omniverse App Streaming feature, access and edit shared virtual worlds with Omniverse Nucleus Cloud, and scale 3D workloads across the cloud with Omniverse Farm.
Omniverse Cloud runs on the planetary-scale Omniverse Cloud Computer. It is powered by NVIDIA OVX for graphics-rich virtual world simulation, NVIDIA HGX for advanced AI workloads, and NVIDIA Graphics Delivery Network to enable low-latency delivery of interactive 3D experiences to edge devices.
Applications for industry workflows—like NVIDIA DRIVE Sim for testing and validating autonomous vehicles and NVIDIA Isaac Sim for training and testing robots—are packaged as containers for simple deployment. For synthetic data for industry use cases, Omniverse Replicator in the cloud enables synthetic 3D data generation for perception networks.
First-class developer experience with Omniverse Kit
Omniverse Kit is a powerful toolkit for building native Omniverse applications and microservices. It is designed to be the premier foundation for new Omniverse-connected tools and microservices.
All the building blocks that developers need for building applications, extensions, and connectors for Omniverse are available in Omniverse Kit. Its modularity enables you to assemble tools in a variety of ways, depending on your unique needs. The Kit team is continuously improving the toolkit with new tools and improved user experience.
Key updates
Viewport 2.0 is now the default, along with many improvements to omni.ui.scene and general availability of the new Viewport menu, enabling you to create your own workflow with Viewport and build your own menu for tools.
Kit Core now supports third-party extensions in C++ and has a number of new features, including Kit Actions for easier scripting and hotkeys, and Kit Activity Monitor for a full timeline of load activity. This means you can easily bring your own C++ library into Kit and build performance-critical code.
Kit Runtime introduces many RTX performance and quality improvements. Action Graph has user interface and user experience improvements across the board, including specific nodes for creating user interfaces.
To get started building quickly, you can take advantage of many documentation improvements, including interactive documentation building and new samples for omni.ui, Scene, and Viewport.
Figure 1. Viewport 2.0 on Omniverse Kit with new menus showing manipulators
Publishing portal on Omniverse Exchange
With the upcoming release of a self-publishing experience for Omniverse Exchange, you as a developer will have a powerful channel to expand the user audience for your connectors, extensions, and asset libraries. The publishing portal will provide a workflow for partners and community members to publish applications, connectors, and extensions to be featured in the Exchange.
Through Omniverse Exchange, Omniverse customers can seamlessly access industry and purpose-built third-party solutions that will accelerate and optimize their workflows. All content undergoes security vetting and quality assurance before being published.
Developers can be among the first to upload extensions or connectors to the NVIDIA Omniverse Exchange Publishing Portal through the early-access program, available by application. Community members are already taking advantage of these developer tools.
Advances to USD
NVIDIA believes that USD is the best candidate to serve as the HTML of the metaverse. USD, an open and extensible ecosystem for describing, composing, simulating, and collaborating within 3D worlds, is now being used in a wide range of industries.
To accelerate the evolution of USD to meet the needs of the metaverse and become the standard language of virtual worlds, NVIDIA is continuing to contribute to the USD ecosystem in all areas. This ranges from education to building custom schemas for specific industry use cases, while providing free USD assets and resources to all audiences.
Key updates for the NVIDIA work in USD
Asset Resolver 2.0 support, enabling Omniverse Connector interoperability with any build of USD
New and improved USD code snippets with documentation for common workflows
USD C++ extension examples for Omniverse Kit, including open-source USD schema examples and tutorials (coming soon)
Support for UTF-8 identifiers for full interchange of content encoded with international character sets (coming soon)
An open-source text render delegate enabling human-readable debugging and more efficient unit and A/B testing of scene delegate implementations (coming soon)
In addition, new Omniverse Extensions extend the functionalities of Omniverse Apps, including Motionverse from Beijing DeepScience Technology Ltd, in3D, and SmartCow. Prevu3D and Move.AI will soon have extensions available, as well.
SimReady assets for simulation workflows
NVIDIA also released SimReady, a new category of 3D asset. 3D assets for digital twins and AI training workloads need specific, USD-based properties. NVIDIA is developing thousands of SimReady assets, as well as specifications for building them, tailored to specific simulation workloads like manipulator bot training and autonomous vehicle training.
SimReady goes beyond stunning, full-fidelity visuals for 3D art assets. It includes content with attached, robust metadata that can be inserted into any Omniverse simulation and behave as it would in the real world. For developers and technical artists, SimReady content can act as a standard baseline or starting point and evolve over time as simulation tools like Omniverse become more robust.
These assets provide full-fidelity visualization with the intent of photorealism. Core simulation metadata is always included in the assets and can be readily accessed on import of the art asset. SimReady assets also leverage the modular nature of USD to provide flexibility for technical artists generating synthetic data.
Video 1. NVIDIA Omniverse Universal Scene Description (USD) SimReady assets
Video 2. USD SimReady Forklift on NVIDIA Omniverse
SimReady asset attributes for developers
Semantic labels to help you train simulation algorithms by identifying the various components of a 3D model in a predictable and consistent way. These labels provide ground-truth arbitrary output variables for annotations within training.
Physical materials that bring simulations closer to reality and enable you to fine-tune how computers see the world with non-visible sensor support like lidar and radar.
Rigid body physics for accurate mass and defined center of gravity to help technical artists create simulations that reflect real-world behavior.
Defined taxonomy with consistent tagging for search and discoverability, so that assets can be usable across many domains.
Robust kinematics and constraints to help you define complex, multi-part relationships and behaviors.
Advanced EM materials to unlock behaviors across the light spectrum.
Download Omniverse to access the first collection of free SimReady assets.
Generate synthetic datasets with Omniverse Replicator
NVIDIA Omniverse Replicator is now deployable in the cloud through containers hosted on NGC and through Omniverse Cloud early access. It features a new Replicator Insight app for enhanced viewing and inspecting of generated data.
Numerous partners, including Siemens SynthAI, SmartCow, Mirage, and Lightning AI, are integrating Omniverse Replicator into their existing tools to extend their synthetic data workflows. Learn more about how Omniverse Replicator is accelerating AI training faster than ever with custom, physically-accurate synthetic data generation pipelines.
Boost rendering performance with Ada Lovelace GPU
Announced at GTC, the new NVIDIA Ada Lovelace GPU delivers real-time rendering, graphics, and AI that will help technical designers and developers build larger virtual worlds. With a more than 2x boost in rendering performance from the previous-generation NVIDIA RTX A6000, NVIDIA Ada Lovelace enables real-time path tracing in 4K resolution. It also enables users to edit and operate large-scale worlds interactively and experience updates to virtual worlds in real time.
A new era of neural graphics
Neural graphics can change the way content is created and experienced. AI-powered tools on NVIDIA AI ToyBox now enable creators to experiment with the latest research in their 3D workflows. AI Animal Explorer is available now, with updates coming soon, including:
GANverse3D with GAN-based and diffusion model-based experimental AI tools
AI Car Explorer with meshed/textured car models by attributes like make and style
Canvas360 with 360-degree landscapes and backgrounds for 3D design
Procedural behavior in Omniverse
Omniverse now enables proceduralism for behavior, movement, and motion in Omniverse worlds. Behavior and physics simulation technologies have improved ease of use, refinement, and functionality. With the new OmniGraph node-based physics systems, behavior, motion, and action can be entirely procedural without any manual animation.
Powerful, flexible simulation with PhysX
NVIDIA PhysX is an advanced real-time physics engine in Omniverse. Several key updates were announced at GTC, including:
Automation improvements and a physics toolbar for object creation
Support for multiple scenes with the ability to put individual assets and stages into their own physics scene
Audio for physics collisions
Ability to author vehicles and force fields in OmniGraph
CPU bottleneck reduction and GPU improvements for soft body simulation
Ability to simulate collisions for particle systems
Expanding ecosystem of extensions with Omniverse Code Contest
This summer, developers from around the world used Omniverse Code to build extensions for 3D worlds as part of the NVIDIA #ExtendOmniverse Contest. Exciting new tools were submitted for layout and scene authoring, Omni.ui, and scene modifier and manipulator tools.
Winners of the contest will be announced at GTC during the NVIDIA Omniverse User Group on September 20 at 3:00 PM (Pacific time).
Explore more at NVIDIA GTC
Join panels, talks, and hands-on labs for developers at GTC to learn more about Omniverse and other tools developers are using to build metaverse extensions and applications.
Watch the latest GTC keynote by NVIDIA founder and CEO Jensen Huang to hear about the latest advancements in AI and the metaverse.
Visit the Omniverse Resource Center to learn how you can build custom USD-based applications and extensions for the platform.
Follow Omniverse on Instagram, Twitter, YouTube, and Medium for additional resources and inspiration. Check out the Omniverse forums, Discord Server, and Twitch channel to chat with the community. Visit NVIDIA-Omniverse GitHub repo to explore code samples and extensions built by the community.
NVIDIA today announced NVIDIA® DLSS 3, an AI-powered performance multiplier that kicks off a new era of NVIDIA RTX™ neural rendering for games and applications.
NVIDIA today unveiled the GeForce RTX® 40 Series of GPUs, designed to deliver revolutionary performance for gamers and creators, led by its new flagship, the RTX 4090 GPU, with up to 4x the performance of its predecessor.
Opening a new era of neural graphics that marries AI and simulation, NVIDIA today announced the NVIDIA RTX™ 6000 workstation GPU, based on its new NVIDIA Ada Lovelace architecture.
Telecoms began touting the benefits of 5G networks six years ago. Yet the race to deliver ultrafast wireless internet today resembles a contest between the tortoise and the hare, as some mobile network operators struggle with costly and complex network requirements. Advanced data analytics company HEAVY.AI today unveiled solutions to put carriers on more even Read article >
With tens of millions of weekly transactions across its more than 2,000 stores, Lowe’s helps customers achieve their home-improvement goals. Now, the Fortune 50 retailer is experimenting with high-tech methods to elevate both the associate and customer experience. Using NVIDIA Omniverse Enterprise to visualize and interact with a store’s digital data, Lowe’s is testing digital Read article >
Deutsche Bahn’s rail network consists of 5,700 stations and 33,000 kilometers of track, making it the largest in Western Europe. Digitale Schiene Deutschland (Digital Rail for Germany, or DSD), part of Germany’s national railway operator Deutsche Bahn, is working to increase the network’s capacity without building new tracks. It’s striving to create a powerful railway Read article >
NVIDIA today announced its first software- and infrastructure-as-a-service offering — NVIDIA Omniverse™ Cloud — a comprehensive suite of cloud services for artists, developers and enterprise teams to design, publish, operate and experience metaverse applications anywhere.
Recent advances in large language models (LLMs) have fueled state-of-the-art performance for NLP applications such as virtual scribes in healthcare, interactive…
Developers, creators, and enterprises around the world are using NVIDIA Omniverse to build virtual worlds and push the boundaries of the metaverse. Based on…