Nvidia's AI Omniverse Expands At GTC 2025

Nvidia's Enhanced Predictability and New Product Lineup at GTC 2025

Nvidia, a key player in AI infrastructure, is focusing on increased predictability for its clients. This involves consistent product delivery, meeting performance expectations, and maintaining reliability. The predictable announcement of the Blackwell Ultra chip (B300) exemplifies this strategy. Simultaneously, Nvidia unveiled Dynamo, a replacement for Triton libraries, designed to accelerate inference, enhance scalability, and reduce deployment costs.

Future Roadmap: Rubin, Vera, and Beyond

Nvidia provided further details on its future roadmap, including the Rubin GPU (early 2026 release with HBM4 memory) and the Vera CPU (2027), part of a broader strategy encompassing compute, NVLink, networking, and system advancements. The Rubin Ultra is slated for later in 2026, and the Vera architecture will underpin the Vera Rubin server series. Nvidia also plans to upgrade NVLink to sixth- and seventh-generation NVSwitches (3600 GB/s bandwidth) paired with Spectrum6 and CX9 networking chips. Spectrum6 is anticipated to be Nvidia's first silicon photonics product. The Feynman GPU, planned for 2028, will utilize "next-gen HBM" and integrate with Vera CPUs, eighth-generation NVSwitch, and Spectrum7 (including its second-generation silicon photonics).

DGX Station and DGX Spark: Accessibility and Scalability

Nvidia launched DGX Spark ("desktop supercomputer"), previously known as Project Digits. This compact device, powered by the GB10 Superchip (combining a Blackwell GPU and a MediaTek 20-core Arm-based CPU), offers accessibility for AI development. Its scalability allows for performance improvements through interconnectivity. Nvidia envisions DGX Spark as an AI appliance for edge computing. Pre-orders were taken during GTC, with ASUS, Dell, and HP also developing GB10-based solutions. Dell and HP showcased workstations based on the DGX Workstation (a single Grace CPU and dual Blackwell GPUs in a liquid-cooled desktop), maximizing power for AI developers. A photo of Huang and Michael Dell signing a Dell GB300 DGX Workstation prototype was captured at GTC 2025.

RTX Pro Series: Professional Graphics Leadership

Nvidia announced the RTX Pro series GPUs, aimed at professional graphics across laptops, desktops, and servers. Building on its AI dominance, Nvidia reinforces its professional graphics leadership. The RTX Pro series (based on GB200 chips) offers five performance levels, with the top-end RTX Pro 6000 boasting a 600-watt, 300 Max-Q configuration. Dell showcased an RTX Pro 6000 workstation. HP and Lenovo are also partners, with HP announcing Nvidia-based systems at its Amplify conference.

The RTX Pro 6000 features 20,064 CUDA cores, 96GB of GDDR7 with ECC, 4,000 TOPS of AI performance (FP4), and 380 TFLOPS of ray tracing performance. VR-SLI is enabled for high-end XR solutions. Laptop versions will range from RTX Pro 5000 (24GB VRAM) to RTX Pro 500. Dell, HP, Lenovo, and Razer will offer RTX Pro laptops.

The author suggests that announcing the RTX Pro series at GTC 2025 may have been premature, suggesting Siggraph 2025 as a more suitable venue.

Nvidia and Cadence Collaboration

Nvidia highlighted its partnership with Cadence, emphasizing the role of Cadence's Palladium emulation platform in Blackwell's development. Nvidia and Cadence are collaborating to accelerate Cadence's design products, including significant performance gains in simulations using Nvidia GPUs. They are also expanding their partnership in datacenter digital twinning using the Cadence Reality platform, integrating with Nvidia Omniverse and OpenUSD. Nvidia is releasing an Omniverse Blueprint for datacenter digital twins.

Nvidia's Continued Growth and Influence

GTC 2025 demonstrated Nvidia's continued expansion and influence in the enterprise, including its growing presence in 5G and 6G networks. The author highlights the impressive performance of the RTX Pro series and concludes that AMD faces a significant challenge in competing with Nvidia's latest offerings.

The above is the detailed content of Nvidia's AI Omniverse Expands At GTC 2025. For more information, please follow other related articles on the PHP Chinese website!