NVIDIA Ships DGX Station, Boosts DGX Spark Agent Performance Up to 2.6x
Updated
Updated · NVIDIA Blog · Jun 1
NVIDIA Ships DGX Station, Boosts DGX Spark Agent Performance Up to 2.6x
7 articles · Updated · NVIDIA Blog · Jun 1
DGX Station deskside supercomputers powered by the GB300 Grace Blackwell Ultra chip are now shipping, with partner systems from ASUS, Dell, GIGABYTE, HP, MSI and Supermicro starting this month.
DGX Spark also got a software update aimed at local AI agents, adding a simplified NemoClaw installer plus support for Hermes Agent and delivering up to 2.6x faster inference on models such as Qwen3.6 35B.
For larger local workloads, NVIDIA's Sync desktop app can now link 2 to 4 DGX Spark systems into a cluster, automatically configuring CX7 networking, SSH access and multinode setup for inference and fine-tuning.
DGX Station targets frontier-scale local development with 748GB of coherent memory and up to 20 petaflops of FP4 performance, enough to run multimodal workflows and models of up to 1 trillion parameters.
The launches fit Jensen Huang's broader COMPUTEX push to make always-on AI agents run across the stack—from cloud AI factories to Windows PCs, deskside systems and robots.
With NVIDIA planning gigawatt-scale 'AI factories,' can the world's energy infrastructure support this AI-powered industrial revolution?
NVIDIA's new 'personal agent' PCs are here. Is this the end of the smartphone's dominance in personal computing?
NVIDIA's toolkit unleashes autonomous AI agents. What new rules are needed to govern a world where software acts independently?
NVIDIA DGX Spark and DGX Station: 2.6x Faster Local AI, 20 PFLOPS Performance, and the Rise of On-Premises Agentic Computing in 2026
Overview
In January 2026, NVIDIA introduced the DGX Spark and DGX Station, bringing powerful, data center-class AI capabilities directly to developers and enterprises. These new systems are designed to enable more efficient and localized AI workflows. The DGX Spark stands out as a pivotal platform, democratizing the ability to run large local AI models. It serves as an exceptional tool for AI developers and content creators by allowing them to offload demanding AI workloads from their personal computers, which helps maintain system responsiveness during content generation. With 128GB of unified memory, DGX Spark supports running large models at full precision.