During his keynote at GTC, Nvidia CEO Jensen Huang introduced the new Ada Lovelace GPU architecture, named after the pioneering mathematician who is considered the first computer programmer. The company also announced two new GPUs based on this architecture: the RTX 6000, aimed at workstations, and the L40, designed for data centers. Alongside these, Nvidia unveiled the second-generation OVX system, powered by the L40 and focused on Omniverse, and introduced Omniverse Cloud, a new software and infrastructure-as-a-service offering.

Ada Lovelace is distinct from Nvidia’s earlier Hopper architecture, which focuses on AI and HPC workloads. Instead, Ada Lovelace is optimized for graphics tasks. While the L40 can handle AI workloads, it is not designed to be as fast as Hopper. It is, however, an excellent universal GPU, particularly strong in graphics performance. The Ada Lovelace architecture is built from the ground up, separate from Hopper, and provides Nvidia with a broader range of solutions for different computing needs.

Both the Ada-generation RTX 6000 and L40 GPUs feature Nvidia’s 4nm custom process, with 76.3 billion transistors, 18,176 CUDA Cores, 568 Tensor Cores, and 142 RT Cores. They also include 48GB of GDDR6 memory. Although both GPUs consume up to 300W of power, the RTX 6000 uses active fan cooling, while the L40 relies on passive cooling. The L40 also supports secure boot with a root of trust. These GPUs are PCIe dual-slot cards but do not use NVLink. The RTX 6000 will be available in a couple of months, with broader availability expected late this year or early next year. The L40 will begin shipping in December.

Nvidia also introduced its second-generation OVX system for Omniverse, which uses the new Ada Lovelace L40 GPU. The updated OVX system is designed to run large Omniverse virtual world simulations, with a massive 48GB frame buffer to support complex digital twin simulations. Each OVX server node is powered by eight L40 GPUs, dual Intel Ice Lake CPUs, and 16TB of NVMe storage, offering powerful processing capabilities. The second-generation OVX systems will be available from several partners, including Inspur, Lenovo, and Supermicro, starting early in 2023.

The L40 GPUs in the OVX system are expected to significantly enhance Omniverse workloads, enabling accelerated ray tracing, photorealistic simulations, and synthetic data generation. In addition to the upgraded GPUs, the OVX system also includes networking enhancements with ConnectX-7 SmartNICs that provide up to 400Gb/s bandwidth and support for 200G networking. This upgrade improves network and storage performance, which is crucial for running precise digital twin simulations.

OVX systems are also highly scalable. From a single pod of eight OVX servers, the system can be expanded into a SuperPOD with 32 servers connected via Nvidia’s Spectrum-3 Ethernet platform. Nvidia emphasized that OVX is flexible, capable of supporting multiple users and different types of applications within Omniverse, making it versatile for various industries.

Among the first customers of the second-generation OVX system are BMW Group and Jaguar Land Rover. Both companies are using the technology to create detailed digital twins of their factories and vehicles, which will help optimize manufacturing processes and improve simulations for autonomous vehicles.

In addition to the new hardware, Nvidia introduced Omniverse Cloud, a suite of cloud-based services that allows artists, developers, and enterprise teams to design, publish, and operate metaverse applications from anywhere. Omniverse Cloud enables collaborative 3D workflows without requiring local computing power, as it leverages Nvidia’s Omniverse Cloud Computer, which includes OVX for graphics and physics simulations and HGX for AI workloads.

Omniverse Cloud offers various services, such as Omniverse App Streaming for users without RTX GPUs, Omniverse Replicator for generating 3D synthetic data, and several other tools like Omniverse Nucleus Cloud, Isaac Sim, and DRIVE Sim. Early supporters of Omniverse Cloud include WPP, Siemens, and RIMAC Group, part of the newly formed Bugatti Rimac company. Omniverse Cloud is available for early access, with users able to deploy certain services via Nvidia’s NGC on AWS using Amazon EC2 G5 instances.