NVIDIA announces a powerful new workstation graphics card, plus its first ever IaaS offering.
NVIDIA’s GTC virtual conference is ongoing this week, and yesterday CEO Jensen Huang delivered his traditional keynote address. It was a flurry of updates and announcements that touched on NVIDIA’s graphics cards, enterprise software, AI products and services, industry partnerships, and much more. There was a little something for everyone from gamers to quantum researchers.
And, of course, engineers—who will soon have access to a powerful new workstation graphics card, the RTX 6000.
The Ada Generation
One of the most exciting announcements for those involved in product design and manufacturing is NVIDIA’s newest architecture for RTX graphics cards, named after pioneering computer scientist Ada Lovelace.
“Ada is phenomenal. It’s a thing of beauty,” said Bob Pette, VP of Professional Visualization at NVIDIA, in a GTC media session.
The Ada Lovelace architecture is the successor to NVIDIA’s previous GPU architecture, Ampere, and according to Pette it’s completely different from the Hopper GPU architecture that NVIDIA announced earlier this year for data center applications. It’s unusual for NVIDIA to have two distinct GPU architectures at the same time, but with so many different use cases for the technology, perhaps it was an inevitable divergence.
The Ada Lovelace architecture will make its debut in the NVIDIA RTX 6000 Ada Generation workstation graphics card, which NVIDIA says provides up to four times the performance of last generation’s Ampere-based RTX A6000. It will be available starting in December.
The NVIDIA RTX 6000 is built with the new third generation of NVIDIA’s RT Cores for real-time ray tracing, the new fourth generation of NVIDIA’s Tensor Cores for machine learning acceleration, and the latest generation of NVIDIA’s long-standing CUDA Cores for shading and other graphics computations. The new graphics card boasts 48GB of GDDR6 graphics memory.
“With the introduction of the NVIDIA RTX 6000 Ada Generation desktop GPU, NVIDIA is providing incredible new levels of performance for professionals to take on the challenges of today and be ready for the evolving, intensive demands of tomorrow,” says Carl Flygare, Professional Graphics and Data Center Product Marketing Manager at PNY.
The new RTX 6000 has a sister graphics card, the NVIDIA L40, also built on the Ada Lovelace architecture. The NVIDIA L40 uses the exact same GPU die as the RTX 6000, but the card itself is designed for data center workloads (for instance, it’s passively cooled while the RTX 6000 has active cooling).
The NVIDIA L40 powers the second-generation of NVIDIA’s OVX server hardware for running real-time graphics and simulations on the Omniverse platform (more on Omniverse later). A single NVIDIA OVX server includes eight L40 GPUs linked with three NVIDIA ConnectX-7 network adapters, and multiple servers can be combined to form what NVIDIA calls an OVX POD (4 to 16 OVX servers) or OVX SuperPOD (32 servers).
In his GTC keynote, Huang also introduced the company’s upcoming consumer graphics cards, the GeForce RTX 4090 and 4080, which are also based on the Ada Lovelace architecture.
Omniverse in the Cloud
Huang’s keynote presentation also showcased several updates to NVIDIA Omniverse, the company’s enterprise software for 3D collaboration and simulation. One noteworthy update is the addition of new Omniverse connectors for Autodesk Alias and the Siemens JT file format, the latter of which Huang considers a big step for the platform.
“Siemens invented JT, the industry standard language of product lifecycle management and the interoperability format of CAD systems like NX, Creo, Catia, and Inventor. The JT connector opens the industrial and manufacturing worlds to Omniverse,” Huang said.
Another big step for Omniverse is that it’s now available in the form of an infrastructure-as-a-service called Omniverse Cloud. This comprises a collection of Omniverse services that can be deployed as containers on Amazon Web Services (AWS)—several of which are currently available—or as a managed service directly from NVIDIA, which is still in early access and taking applications.
Omniverse Cloud services include Omniverse Nucleus Cloud for data management and synchronization, Omniverse App Streaming for building and viewing projects, Omniverse Replicator for generating synthetic training data, Omniverse Farm for scaling-out workloads, and NVIDIA Isaac Sim and DRIVE Sim for simulating robots and autonomous vehicles, respectively.
What kind of computer will run Omniverse Cloud? The Omniverse Cloud Computer, of course.
“Omniverse is a new computing platform, and requires a new computer system,” Huang said. The “planetary-scale” Omniverse Cloud Computer consists of OVX servers for virtual world simulations, HGX servers for AI workloads, and what NVIDIA is calling its Graphics Delivery Network (GDN) to deliver 3D content to edge devices around the globe (“portals into the Omniverse,” in Huang’s words).
The Omniverse userbase continues to grow with every GTC. “When it comes to creating and connecting virtual worlds, over 150,000 individuals have downloaded NVIDIA Omniverse to make huge leaps in transforming 3D design workflows and achieve new heights of real-time, physically accurate simulations,” Flygare says.
Huang highlighted several companies that have begun to use Omniverse. Hardware retailer Lowe’s is using the platform to create digital twins of its stores, and automaker Rimac is using Omniverse Cloud to build and deploy a 3D product configurator for its Nevera electric sports car.