New systems, processors and partnerships simplify the path to generative AI.
At this year’s SIGGRAPH 2023 conference in Los Angeles, NVIDIA made a series of announcements that highlight the company’s innovations, particularly in the field of generative artificial intelligence (AI) and graphics.
Five years ago, NVIDIA introduced RTX, enabling real-time ray tracing with graphics processing units (GPUs) through AI. Not long after, it launched GauGAN, an early form of generative AI that can translate doodles and text into images.
Today, Nvidia is customizing and fine-tuning generative AI models in applications ranging from text-to-image and text-to-3D.
Generative AI: Easy as 1-2-3
NVIDIA follows a three-step process that involves utilizing the right foundation models, customizing them with domain-specific data, and incorporating them into apps and services for production.
During a SIGGRAPH news briefing, NVIDIA’s VP of enterprise computing, Manuvir Das, emphasized the importance of customizing pre-trained models for specific apps. To demonstrate how the models worked, he gave a command to create “Toy Jensen” in space.
As expected, the more images the model was fed, the more accurate the results. What was a surprise was that it took only eight images to create a workable image, and then by 50 it was fully accurate.
“This is a message that NVIDIA has been delivering since the beginning of this year in every forum. The real-life example of Toy Jensen is where we used our toolkit to fine-tune it with fifty images of Jensen, highlighting the efficacy and the output of the model,” Das said.
NVIDIA Partners with Hugging Face for Generative AI
To help provide companies with starting points for working on generative AI models, NVIDIA formed a partnership with a platform called Hugging Face.
At SIGGRAPH, NVIDIA announced that Hugging Face is adding a new service to its site, allowing users to train and fine-tune AI models. Hugging Face will integrate the training service on NVIDIA’s DGX Cloud, housing all the necessary software for customization, guidance, and inference.
Also, NVIDIA unveiled AI Enterprise 4.0, a major upgrade to its enterprise-level AI software.
AI Enterprise 4.0 now integrates NVIDIA NeMo, providing organizations with a toolkit to process generative AI. NeMo allows developers to build, customize, and deploy large-scale models with billions of parameters. It also supports multi-GPU and multi-node configurations, streamlining the process of handling extensive models.
NVIDIA: AI Workbench Makes AI Location Independent
NVIDIA’s third announcement at SIGGRAPH centered on flexibility and mobility with the introduction of the NVIDIA AI Workbench, designed to run directly on a user’s laptop.
Users can select the location where they want to do their work, whether on a PC, data center, cloud, or NVIDIA’s DGX Cloud. Moving between these different locations is designed to be a one-click process, all packaged up within the AI Workbench.
“Whether you’re an independent software vendor (ISV) or a data scientist, AI Workbench is your single pane of glass. It’s a way for you to package up your AI work uniformly and consistently and move it from one location to another. So, you do your work one time, the same way, no matter where you are,” Das said.
More Updates to Omniverse
Additionally, NVIDIA shared updates on the continued growth of its Omniverse platform, which is reshaping industries such as manufacturing, transportation, and telecommunications.
Omniverse is a tool that lets artists and designers work together on 3D projects. It helps create realistic images and animations by utilizing Universal Scene Description (OpenUSD). NVIDIA has formed major partnerships with brands like Denza, WPP, Pixar, Adobe, Apple, and Autodesk, standardizing OpenUSD and transforming traditional design processes.
For example, Denza, a joint venture between BYD and Mercedes-Benz, has tapped WPP to build a fully-interactive 3D car configurator on Omniverse Cloud.
“To build a configurator today, you need to individually render hundreds of thousands of 2D images with every possible variant of the car. With Omniverse, WPP and BYD can create a single super digital twin of a car model based off of all the original engineering and design data, and deploy it across marketing channels instantaneously without any manual rework,” said Rev Lebaredian, VP of Omniverse and simulation technology at NVIDIA.
NVIDIA also made a major update to Omniverse, enhancing its 3D graphics rendering capabilities and adding a centralized tool that simplifies the management of Omniverse’s features for developers. Furthermore, the update provides pre-designed OpenUSD templates, making the creation of 3D apps more streamlined.
New NVIDIA Workstations, GPUs
NVIDIA launched new high-end RTX workstations and three new GPUs: RTX 5000 (available now), RX 4500 (available in October), and RTX 4000 (available in September).
The GPUs are set to revolutionize photorealistic ray tracing and AI video enhancements, solidifying NVIDIA’s position as an innovator in GPU technology, said Bob Pette, VP of Professional Visualization at NVIDIA.
“RTX 4000 is going to be the most powerful single slot GPU on the planet, with massive breakthroughs, speed, and power efficiency—packing this level of performance into a single card,” Pette said.
Lastly, at the event, NVIDIA announced its next-generation NVIDIA GH200 Grace Hopper platform based on the new HMB3e Grace Hopper Superchip, explicitly built for the era of generative AI.
The processor is designed to handle generative AI workloads such as large language models, recommender systems, and vector databases. The new platform will be available in several configurations for various requirements. The dual configuration delivers 3.5x more memory capacity and 3x more bandwidth than the current generation.
Bottom Line: Making Generative AI More Accessible and Versatile
At the SIGGRAPH conference, NVIDIA showcased its efforts to make generative AI not only more accessible but also more versatile. Through strategic partnerships and technological advancements, NVIDIA’s innovations echo a significant industry trend where many different sectors are beginning to harness the power of AI.
The AI space has become highly competitive, but NVIDIA has maintained and arguably stretched its lead over rivals AMD and Intel.
What makes the company unique is that it takes a systems approach. Many think of NVIDIA as a GPU manufacturer but, in reality, it delivers not only the processors but the programming languages, platforms, systems and partnerships required to simplify the journey to AI.