XpertStation WS300: the AI supercomputer for your desk
NVIDIA Grace Blackwell Ultra

A data center, built to sit beside your desk.

XpertStation WS300 brings NVIDIA DGX Station architecture to the desktop. Develop, fine-tune and serve large AI models on local data, with 748 GB of coherent memory in a single tower.

  • 748 GBCoherent memory
  • 800 Gb/sConnectX-8 SuperNIC
  • 1600 W80 PLUS Platinum

What sets the XpertStation WS300 apart

Data-center-class architecture, scaled to a single tower and built around one coherent memory pool.

Compute-dense architecture

The NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip joins a high-performance NVIDIA Grace CPU and a Blackwell Ultra GPU over the NVLink Chip-to-Chip (C2C) interconnect. The result is exceptional compute density and efficient communication across the system.

High-speed networking

The NVIDIA ConnectX-8 SuperNIC delivers up to 800 Gb/s of low-latency, high-bandwidth connectivity, for fast data transfer and scalable performance on large AI workloads.

Large coherent memory

A 748 GB coherent pool combines 252 GB of HBM3e GPU memory with 496 GB of LPDDR5X system memory through the C2C interconnect. Data is reached faster, and large-scale model training runs more efficiently.

End-to-end AI software support

The NVIDIA AI software stack covers fine-tuning, inference and data science, so work moves from desktop to data center without starting over.

Accelerating every AI workload

XpertStation WS300 on NVIDIA DGX Station architecture supports model development, data science and autonomous AI agents with NVIDIA OpenShell.

AI development

Speed up deep learning and machine learning training for applications from predictive maintenance and medical imaging analysis to natural language processing.

Data science

Accelerate end-to-end workflows for faster data ingestion, analysis and insight across massive datasets.

AI inference

Run large, complex models locally with high-speed LLM token generation, data analysis, content creation and AI chatbots.

Personal cloud

Run advanced models on local data, and serve as a central compute node for team fine-tuning and on-demand deployment.

Specifications

Everything in the chassis, from superchip to power supply.

Superchip
NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip(1) NVIDIA Blackwell Ultra GPU
CPU memory
Up to 496 GB LPDDR5XUp to 396 GB/s
GPU memory
Up to 252 GB HBM3e7.1 TB/s
Networking
(1) 10 GbE RJ45 port, Marvell AQC113(2) 400 GbE QSFP ports, NVIDIA ConnectX-8
Storage
(2) 2280 PCIe 5.0 x2 NVMe M.2 ports, from CPU(2) 2280 PCIe 6.0 x4 NVMe M.2 ports, for training data
Expansion
(1) double-wide PCIe 5.0 x16 slot for display(2) single-wide PCIe 5.0 x16 slots, each with x8 signal only
(1) PCIe x1 2232 M.2 for Wi-Fi/Bluetooth card
Management
Qualified BMC module, Aspeed AST2600
Power
1600 W 80 PLUS Platinum ATX PSU
XpertStation WS300 on NVIDIA DGX Station architectureNVIDIA, Grace Blackwell, NVLink and ConnectX are trademarks of NVIDIA Corporation.
Shopping cart

Your cart is empty.

Return to shop