HP Opens Orders for AI Workstation 'ZGX Fury' with GB300 Superchip

HP has started taking orders for the ZGX Fury, an enterprise AI workstation powered by NVIDIA's GB300 Grace Blackwell Ultra Desktop Superchip. The product is positioned not as a personal workstation, but as a shared inference box for use in departments or on-site without a data center.

Hardware Configuration and Performance

• Processor: One Blackwell Ultra GPU (252GB HBM3e, 7.1TB/s bandwidth) and a 72-core Grace CPU (496GB LPDDR5X) are connected via NVLink-C2C.
• Memory: Provides a total 748GB of coherent memory space. With FP4 quantization, it enables 1 trillion parameter inference and fine-tuning of models with 100 billion parameters in a single box.
• Storage and Networking: Supports two PCIe 5.0 M.2 slots for the OS and two NVMe slots for data, and can connect to other ZGX Fury systems via two 400Gbps QSFP112 ports through ConnectX-8.
• Cooling and Design: Offers tower and 5U rack mount options with liquid cooling and optimized airflow. It has no display output and is designed exclusively for inference.

Software and Ecosystem Integration

• Red Hat Partnership: Has completed Red Hat Enterprise Linux certification and applies the Red Hat AI Factory with NVIDIA platform to support model and agent deployment in hybrid cloud environments.
• HP-Specific Tools: Comes with HP Z Runtime for local model management and the open-source framework HP Z Toolkit as standard. It supports a workflow of prototyping on the ZGX Nano and scaling up to the ZGX Fury.
• Operating System: Ubuntu 24.04 LTS and NVIDIA AI development tools are pre-installed.

Market Situation and Availability

• Currently available for order through HP, but pricing has not been disclosed. The Red Hat AI Factory integrated solution is a planned offering, with specific support configurations and timing to be announced later.
• Competition is expected from products like MSI's WS300 and AMD's Threadripper Halo Station, which target similar workloads based on the GB300 superchip.