NVIDIA unveiled the DGX Station and DGX Spark at GTC 2025. DGX Station brings data-center-class AI performance to your desktop workspace. Powered by the Grace Blackwell Ultra desktop superchip, it offers up to 784 GB of unified coherent memory—enabling developers, researchers, and data scientists to train and fine-tune large AI models locally, without needing a full rack. It’s a true personal AI supercomputer, designed for maximum innovation in a compact form.
Advanced Architecture for Extreme AI Performance
This system is built around NVIDIA’s GB300 Grace Blackwell Ultra chip, which tightly integrates the Blackwell GPU and Grace CPU via NVLink-C2C for ultra-high bandwidth communication. It features 5th-gen Tensor cores, enabling 4-bit FP4 AI precision for large, efficient model training. With the ConnectX-8 SuperNIC, it supports up to 800 Gb/s networking, making it easy to scale or connect multiple units seamlessly.
Optimized for Real-World AI Development and Scalable Team Workflows
DGX Station excels in prototyping, fine-tuning, and inference of large language models and data-intensive AI applications. Its 7-way Multi-Instance GPU (MIG) support enables collaborative workloads: multiple users can run isolated GPU instances concurrently. Whether serving a small research team or powering a dev-ops workflow, DGX Station delivers flexibility, performance, and enterprise-grade software support.
Blackwell Ultra Chip and 5th Gen Tensor Cores
NVIDIA Grace Blackwell Ultra Desktop Superchip
The NVIDIA DGX Station comes with a NVIDIA Blackwell Ultra GPU, built with the latest generation NVIDIA CUDA® cores and 5th generation Tensor Cores, connected to a high-performance NVIDIA Grace CPU via the NVIDIA® NVLink®-C2C interconnect, delivering best-in-class system communication and performance.
5th Generation Tensor Cores
NVIDIA DGX Stations are powered by the latest NVIDIA Blackwell Generation Tensor Cores, enabling 4-bit floating point (FP4) AI. FP4 increases the performance and size of next-generation models that memory can support while maintaining high accuracy.
Workload Optimized Power-shifting
NVIDIA DGX Stations take advantage of AI-based system optimizations that intelligently shift power based on the currently active workload, continually maximizing performance and efficiency.
Networking and Interconnect
NVIDIA ConnectX-8 SuperNIC
The NVIDIA ConnectX®-8 SuperNIC™ is optimized to supercharge hyperscale AI computing workloads. With up to 800 gigabits per second (Gb/s), NVIDIA ConnectX-8 SuperNIC delivers extremely fast, efficient network connectivity, significantly enhancing system performance for AI factories.
NVIDIA NVLink-C2C Interconnect
NVIDIA NVLink-C2C extends the industry-leading NVLink technology to a chip-to-chip interconnect between the GPU and CPU, enabling high-bandwidth coherent data transfers between processors and accelerators.
Large Coherent Memory for AI
AI models continue to grow in scale and complexity. Training and running these models within NVIDIA Grace Blackwell Ultra’s large coherent memory allows for massive-scale models to be trained and run efficiently within one memory pool, thanks to the C2C superchip interconnect that bypasses the bottlenecks of traditional CPU and GPU systems.
Software Optimized for AI and ML
NVIDIA AI Software Stack
A full stack solution for AI workloads including fine-tuning, inference, and data science. NVIDIA’s AI software stack lets you work local and easily deploy to cloud or data center using the same tools, libraries, frameworks and pretrained models from desktop to cloud.
NVIDIA DGX OS
NVIDIA DGX OS implements stable, fully qualified operating systems for running AI, machine learning, and analytics applications on the NVIDIA DGX platform. It includes system-specific configurations, drivers, and diagnostic and monitoring tools. Easily scale across multiple NVIDIA DGX Station systems, NVIDIA DGX Cloud, or other accelerated data center or cloud infrastructure.
NVIDIA DGX Station Specifications
NVIDIA GPU 1x NVIDIA Blackwell Ultra
NVIDIA CPU 1x Grace-72 Core Neoverse V2
GPU Memory Up to 288GB HBM3e | 8 TB/s
CPU Memory Up to 496GB LPDDR5X | Up to 396 GB/s
NVLink-C2C Up to 900 GB/s
Networking | Peak Bandwidth NVIDIA ConnectX®-8 SuperNIC | Up to 800 Gb/s
Supported OS NVIDIA DGX OS
MIG 7
