Five years ago at SIGGRAPH, NVIDIA reinvented graphics by incorporating AI and real-time ray tracing into its GPUs. Jensen Huang said, "We were reinventing computer graphics with AI, while also completely redesigning the GPU specifically for AI."

The result is increasingly powerful systems like the NVIDIA HGX H100, which leverages eight GPUs and a total of one trillion transistors to deliver breakthrough acceleration over CPU-based systems.

To continue the momentum in AI, NVIDIA developed the NVIDIA GH200, a Grace Hopper superchip that combines a 72-core Grace CPU and a Hopper GPU, and began full-scale production in May.

Jensen Huang announced that the NVIDIA GH200, which is already in production, will be complemented by an additional version equipped with cutting-edge HBM3e memory.

He then announced the next-generation GH200 Grace Hopper superchip platform, which supports server designs that connect multiple GPUs for high performance and easy scalability. The new platform is designed to handle the world's most complex generative workloads, spanning large-scale language models, recommender systems, and vector databases, and will be available in a variety of configurations.

Delivering up to 3.5x more memory capacity and 3x more bandwidth than current-generation products, the dual configuration features a single server with 144 Arm Neoverse cores, 8 petaflops of AI performance, and 282GB of the latest HBM3e memory technology.

Major system manufacturers are expected to offer systems based on this platform in the second quarter of 2024.

NVIDIA AI Workbench Accelerates Adoption of Custom Generative AI
Jensen Huang announced the AI Workbench, which provides developers with an easy-to-use, integrated toolkit that enables them to quickly create, test, and fine-tune generative AI models on their PC or workstation, then scale them to virtually any data center, public cloud, or NVIDIA DGX Cloud.

The AI Workbench removes the complexity that arises in the early stages of enterprise AI projects. Accessible through a streamlined interface that runs on local systems, developers can fine-tune models using custom data from popular repositories like Hugging Face, GitHub, and NGC. Models can be easily shared across multiple platforms.

While hundreds of thousands of pre-trained models are available today, customizing them using open-source tools can be challenging and time-consuming.

The AI Workbench allows developers to customize and run generative AI with just a few clicks. It brings all the necessary enterprise-grade models, frameworks, software development kits, and libraries into a unified developer workspace.

Leading AI infrastructure providers, including Dell Technologies, Hewlett Packard Enterprise, HP Inc., Lambda, Lenovo, and Supermicro, are adopting AI workbenches that provide enterprise-ready AI capabilities wherever developers want, including on local devices.

Jensen Huang also announced a partnership with Hugging Face, a startup with 2 million users. This will make generative AI supercomputing readily available to millions of developers building large-scale language models and other advanced AI applications.

In the video, he demonstrated how AI Workbench and ChatUSD bring everything together so that users can start a project on a GeForce RTX 4090 laptop and seamlessly scale to a workstation or data center as the project grows in complexity.

A user can use Jupyter Notebook to input a message to the model, asking it to generate a photo of Toy Jensen in space. If the model fails to produce a result because it has never seen Toy Jensen, the user can fine-tune the model with eight images of Toy Jensen. You can then request the message again to ensure you get the correct result.

This new model can then be deployed into enterprise applications via the AI workbench.

■ A new omniverse that combines generative AI and OpenUSD is launched.

Jensen Huang announced a new version of NVIDIA Omniverse, the OpenUSD native development platform for building, simulating, and collaborating across tools and virtual worlds. This will enable developers and enterprises to deliver new foundational applications and services that optimize and enhance their 3D pipelines with the OpenUSD framework and generative AI.

Updates to the Omniverse platform include the Omniverse Kit, an engine for developing native OpenUSD applications and extensions, the NVIDIA Omniverse Audio2Face Foundation app, and improved performance for spatial computing capabilities.

Cesisium, Convai, Move AI, SideFX Houdini, and Wonder Dynamics are now connected to Omniverse via OpenUSD.

Additionally, Adobe and NVIDIA announced plans to expand their collaboration across Adobe Substance 3D, Generative AI, and the OpenUSD initiative, and to make Adobe Firefly, Adobe’s suite of creative generative AI models, available as an API in Omniverse.

Omniverse users can now create content, experiences, and applications that are compatible with other OpenUSD-based spatial computing platforms, such as ARKit and RealityKit.

Jensen Huang announced a comprehensive framework, resources, and services to accelerate the adoption of Universal Scene Description (OpenUSD) by developers and enterprises. This includes geospatial data models, metrics assembly, simulation support, and SimReady.

He also announced four new Omniverse Cloud APIs built by NVIDIA. This will enable developers to more seamlessly implement and deploy OpenUSD pipelines and applications.

■ New desktop systems and servers

NVIDIA and global workstation manufacturers have announced new RTX workstations, delivering more powerful computing to support development and content creation in the era of generative AI and digitalization.

The system, which will be deployed in BOXX, Dell Technologies, HP, and Lenovo, is based on the NVIDIA RTX 6000 Ada Generation GPU and integrates NVIDIA AI Enterprise and NVIDIA Omniverse Enterprise software.

Additionally, NVIDIA has launched three new Ada-generation GPUs for desktop workstations: the NVIDIA RTX 5000, RTX 4500, and RTX 4000, to deliver the latest AI, graphics, and real-time rendering technologies to professionals around the world.

NVIDIA continues to accelerate generative AI and industrial digitalization with global data center system manufacturers through the new NVIDIA OVX, featuring the new NVIDIA L40S GPU, a powerful general-purpose data center processor design.

These powerful new systems, powered by the NVIDIA Omniverse platform, accelerate the most compute-intensive and complex applications, including AI training and inference, 3D design and visualization, video processing, and industrial digitization.