NVIDIA has unveiled its DGX Station, which lets users develop and run artificial intelligence (AI) models with up to 1 trillion parameters locally on Windows. The DGX Station expected in Q4 this year will let businesses build and deploy their own AI without sending their data to an external cloud, effectively giving enterprises a desk-sized AI supercomputer in-house. 

In the short history of AI deployment by businesses, the tasks of training, fine-tuning, large-scale inference, and more have relied on powerful AI systems running on Linux. However, businesses do not run on Linux. Their productivity tools, design, or engineering applications all on Windows. 

Developing AI tools for internal use, therefore, requires an additional step: making data accessible to AI systems that run on Linux. With the DGX Station for Windows, NVIDIA is removing this complexity and making it easier to build, run, and connect AI agents on existing apps and infrastructure on the Windows platform. 

What is the DGX Station? 

The DGX Station is powered by NVIDIA’s GB300 Grace Blackwell Ultra Desktop Superchip that connects with NVIDIA’s 72-core Grace CPU. The system boasts 748 GB of coherent memory with up to 20 petaflops of FP4 performance. 

The DGX Station features NVIDIA’s ConnectX-8 SuperNIC, which supports 800 Gbps networking optimized for hyperscale AI computing workloads. The ConnectX-8 SuperNIC can be deployed to enable extremely fast data networks for AI workloads or to interconnect multiple DGX stations, delivering even more powerful solutions. 

“As enterprises scale AI agents across their organizations, they need AI infrastructure that can connect directly to the applications and workflows that power their business,” said Chris Marriott, vice president of enterprise platforms at NVIDIA, in a press release. 

“DGX Station delivers supercomputing-class AI directly into Windows, where millions already design, engineer, research and create every day,” added Marriott.  

Developed in collaboration with Microsoft, the DGX Station serves as a dedicated agent infrastructure for running AI models with up to 1 trillion parameters. If required, the Station can be configured to run hundreds of AI agents simultaneously at scale. 

Always On AI agents 

While our experiences with AI agents might be of those that respond to our queries when prompted, businesses’ use of AI runs differently. Enterprise AI is always on, connected to applications and workflows, as it continues to reason and respond in real time.

With large-scale access to data, autonomous agents need to be developed and deployed in a secure runtime that governs how they act, which tools they can use, and their role in a larger system.

DGX aims to provide companies with a secure runtime for deploying and testing AI agents before scaling to data centers. NVIDIA’s OpenShell provides a secure runtime environment and leverages new Windows security features to create individual isolated sandboxes  for AI agents to operate in, 

By separating application-layer operations from infrastructure-layer policy enforcement, the Station puts privacy and security policies outside the agent’s reach to avoid them being leaked or overridden. 

“For decades, Microsoft and NVIDIA have partnered to advance the most powerful computing platforms in the world,” said Pavan Davuluri, executive vice president of Windows + Devices at Microsoft in a press release. 

“Today, we’re taking that collaboration to the next level, scaling the full power of Windows from thin-and-light PCs to data-center-class workstations with DGX Station powered by GB300. This unlocks a new class of AI performance on Windows, the platform enterprises trust for security, manageability and compatibility.”

Recommended Articles