Nvidia 在 GTC Berlin 发布 DSX POD 等 AI 参考架构与开发工具

Nvidia Corp. today debuted a set of technologies that will make it easier for customers to build data centers and robots using its chips.

Chief Executive Officer Jensen Huang detailed the offerings at the company’s annual GTC Berlin event.

The first addition to Nvidia’s data center portfolio, DSX POD, is a reference architecture. That’s a kind of blueprint partners can use to create products based on its silicon. DSX POD enables manufacturers to build artificial intelligence clusters that fit in containers. Each container hosts Rubin graphics processing units, Vera central processing units, network equipment, cooling gear and power distribution modules.

Nvidia says the components in DSX POD containers are integrated and tested at the factory. That removes the need for customers to assemble the systems, which can significantly speed up AI cluster construction. Nvidia estimates that DSX POD can cut hardware deployment times from six months to 14 days.

Dell Technologies Inc. and Hewlett Packard Enterprise Co. are among the more than dozen hardware makers that plan to adopt DSX POD. Nvidia expects the first systems to ship early next year.

“An AI factory creates value when it starts producing tokens — not when construction is complete,” Jensen Huang said.

The company debuted DSX POD alongside a second data center reference architecture. The latter blueprint is designed to help customers deploy OpenShell, an AI agent toolkit that Nvidia released last month. The software runs AI agents in sandboxes that isolate them from sensitive infrastructure. A supervisor module monitors the traffic that exits sandboxes for cybersecurity risks.

The reference architecture makes it easier to run OpenShell on Nvidia silicon. It combines Vera CPUs, Rubin GPUs and BlueField-4 accelerators. The latter chips offload certain auxiliary computing tasks from the other processors in a data center.

The new blueprint also enables users to run OpenShell environments at the edge. It supports Nvidia’s DGX Station workstations, DGX Spark miniature desktop and laptops that feature its RTX Spark system-on-chip.

“As enterprises prepare to deploy billions of AI agents, safety and control must be built into the computing architecture,” Huang said. “Nvidia created OpenShell to give agents the freedom to do useful work within clearly defined boundaries — contained, monitored and under control.”

The third major focus of Huang’s keynote was the robotics market. Nvidia offers a software toolkit called Isaac that developers can use to build AI models for robots. It includes prepackaged foundation models, a robot simulator and several related tools.

Nvidia launched a new version of Isaac’s simulator that features an enhanced set of application programming interfaces. Developers can use them to define details such as the sensors that a virtual robot should use to navigate a simulated factory. The update is rolling out alongside three template workflows for customizing GR00T, a robot-optimized foundation model included in Isaac.

GR00T powers Agile ONE SN, a 5.6-foot-tall humanoid robot that also made its debut today. The system was developed by Agile Robotics Inc., a startup that counts Nvidia as an investor. The chipmaker says that customers can use Agile ONE SN as the starting point of humanoid robot projects. The system takes less time to set up than the hardware that engineers typically use for the task.

Huang’s GTC Berlin keynote placed an emphasis on the role Nvidia products are playing in the European Union’s AI ecosystem. He detailed that Mercedes-Benz will use the chipmaker’s Hyperion computing and sensing module to build autonomous vehicles. Deutsche Telekom AG, meanwhile, plans to expand its Industrial AI Cloud, an Nvidia-powered service that hosts open-source frontier models for customers.

“As expected, Jensen positioned Nvidia as a strong supporter of — and enabler for — Europe’s sovereign tech ambitions,” said Forrester Vice President and Principal Analyst Paul Miller. “Nvidia has invested almost $10bn into Europe in the past year, including 14 investments in European startups during Jensen’s current trip.”

Photo: Nvidia