Nvidia Announces New and Expanded Products at SIGGRAPH 2025

The SIGGRAPH 2025 computer graphics show takes place this week in Vancouver, and Nvidia has already made several significant announcements. Given that computer graphics is one of Nvidia's core strengths, SIGGRAPH has always been a major event for the company, and this year was no exception. The company announced several new products and expanded existing capabilities across its fast-growing portfolio of AI offerings.

Below are the key highlights among the many announcements at the show.

RTX PRO Servers: Breakthrough Performance

NVIDIA unveiled two new RTX PRO servers that will be incorporated into enterprise systems from companies such as Cisco, Dell, HPE, Lenovo and Supermicro. These new servers -- powered by the RTX PRO 6000 Blackwell GPU -- deliver "breakthrough performance versus the previous generation of L40S," said Rev Lebaredian, vice president of omniverse and simulation technology at Nvidia, in an analyst briefing.

According to Nvidia, the new servers deliver the following:

While I can't independently verify these claims, such performance leaps are consistent with historical Nvidia innovation.

Related:4 Takeaways from Antonio Neri's Keynote at HPE Discover 2025

With the increasing demand for AI , computer graphics and other advanced computing applications, these servers can help customers accelerate the shift from traditional CPU-based systems to GPU-based accelerated computing platforms. A new RTX PRO server configuration for high-volume mainstream systems features two RTX PRO 6000 GPUs in a 2U form factor for rack-mounted servers. These new servers deliver 45x the performance for applications such as data analytics, simulation and rendering, resulting in 8x higher energy efficiency and lower total cost of ownership compared to CPU-only systems.

In previous keynotes, Nvidia CEO Jensen Huang joked that with accelerated computing platforms, "the more you buy, the more you save." While his comments are slightly tongue-in-cheek, the reality is a single RTX PRO replaces multiple racks of CPU-based systems.

New Nemotron Training Sets: Enhancing AI Reasoning

Nvidia has added two new data sets to its Nemotron family of multimodal models to empower AI agents to "reason better and make smarter decisions," Lebaredian said. A new hybrid model architecture and configurable thinking budget feature give agents the ability to think more deeply and work more efficiently. This helps agents explore broader options and deliver smarter results within set time boundaries.

Related:2 Ways to Think about AI and Networking -- Without the Hype

As AI evolves toward agentic systems, reasoning capabilities become critical for making decisions on behalf of humans. Many AI experts predict a rise in digital twins in which an AI version of everyone can complete complex tasks for them. These new models are a step in that direction.

According to Nvidia, the Nemotron Nano 2 can generate up to 6x more tokens in a given timeframe and lower reasoning costs by 60% compared to other models in its class. The Llama Nemotron Super 1.5 now supports NVFP4 on Blackwell, "delivering 6x throughput compared to FP8 on Hopper," Lebaredian said.

Both new Nemotron data sets will be available for download soon from Nvidia and Hugging Face.

Improving Physical AI

When many people think of AI, they probably first envision software that writes documents, answers questions and handles repetitive tasks. But Nvidia's vision goes far beyond that. The company is investing in numerous physical AI initiatives to help organizations that deploy physical robots ensure they interact safely with humans.

One of Nvidia's major initiatives in this space is Metropolis , a vision AI application platform and partner ecosystem that simplifies the development, deployment and scaling of visual AI agents across various use cases.

Smart cities, facilities and industrial processes are just some of the current and future applications of Metropolis. Nvidia works with a range of organizations deploying Metropolis to develop new use cases -- from enhancing safety for workers who operate heavy machinery to improving public transportation and supporting public safety organizations.

Key Metropolis Updates

1. Cosmos Reason

Cosmos Reason is key to Nvidia's physical AI offering. It's an advanced, open, customizable, 7-billion-parameter reasoning vision-language model (VLM). The latest version enables contextual video understanding and temporal event reasoning for Metropolis use cases. Because of its small footprint, organizations can deploy Cosmos Reason from the edge to the cloud for tasks such as traffic monitoring, visual inspection and intelligent decision-making.

2. VSS Blueprint 2.4

The second update is video search and summarization (VSS) Blueprint 2.4, which simplifies the process of augmenting existing vision AI applications with Cosmos Reason and delivering new features to smart infrastructure. An expanded set of APIs in the blueprint gives users more flexibility when choosing specific VSS components and capabilities to enhance computer vision pipelines with generative AI.

3. Isaac Sim Extensions

New extensions in Nvidia's Isaac Sim reference application are designed to help resolve AI development issues. By simulating human and robot interactions, the tools generate rich object-detection datasets and train VLMs by creating and implementing incident-based scenes and image-caption pairs. This speeds up development and improves real-world AI performance.

4. Additional Hardware Support

Metropolis components now run on Nvidia RTX Pro 6000 Blackwell GPUs, the Nvidia DGX Spark desktop supercomputer and Nvidia's platform for physical AI and humanoid robots -- Jetson Thor. This enhances users' ability to develop and deploy from the edge to the cloud.

Cosmos Reason and Nvidia TAO 6.0 -- for fine-tuning pretrained models with company-specific data -- are available now for download.

Final Thoughts

AI comes up in any event or conversation I have with executives -- and it's typically the main topic. Nvidia has long pushed the boundaries of accelerated computing, but the use cases were limited to graphics and scientific analysis. Today, AI is everywhere, making Nvidia's innovations critical to the future of ubiquitous AI.