Skip to content
Breaking:

HP Introduces $132,000 ZGX Fury AI Workstation Built on Nvidia GB300

The liquid-cooled system delivers 748GB of split unified memory, targeting engineering teams running local language models without cloud dependencies.

By The Company Wire3 min read
Share
HP — HP Introduces $132,000 ZGX Fury AI Workstation Built on Nvidia GB300
Rendering. An HP ZGX Fury workstation tower featuring front I/O ports and a mesh ventilation panel against a black background. Photo: TechRadar Pro

HP has unveiled the ZGX Fury G1n, an enterprise-grade desktop workstation designed to run demanding artificial intelligence workflows directly inside research labs and corporate offices. Starting at $132,000, the machine integrates Nvidia's GB300 Grace Blackwell Ultra platform, bringing compute hardware typically reserved for hyperscale data centers into an office enclosure, according to reporting by TechRadar Pro (https://www.techradar.com/pro/your-next-desktop-pc-for-the-price-of-a-ferrari-hp-zgx-fury-packs-in-748gb-ram-and-nvidia-gb300-but-starts-at-usd132-000).

The launch addresses enterprise demand for local model iteration, where software teams look to evaluate and adapt large language models without recurring cloud compute bills or the compliance challenges of transferring proprietary datasets to hosted platforms.

The workstation's architecture splits 748GB of total memory across two high-speed tiers. The primary system pool features 496GB of LPDDR5X memory running at 6,400 MT/s wired to a 72-core Arm Neoverse V2 Grace CPU. Dedicated graphical and accelerator memory consists of 252GB of high-bandwidth HBM3e directly integrated into the Blackwell Ultra GPU. HP also supports configurations with discrete Nvidia RTX PRO workstation cards—including the RTX PRO 2000 and RTX PRO 4000—which can push dedicated video memory to 276GB.

HP states that this memory footprint allows teams to locally fine-tune language models containing up to 100 billion parameters. Using FP4 quantization, the company claims the hardware can execute local inference on models scaling to one trillion parameters, though real-world throughput varies based on specific model architectures and operational constraints.

Dissipating the thermal load of the dense compute cluster requires specialized internal plumbing. HP configured the machine with closed liquid cooling backed by a 1,600-watt Titanium-certified power supply. The physical chassis can sit deskside or mount directly into a standard 5U server rack for clustered deployments.

To integrate with existing distributed training infrastructure, the system ships with Ubuntu 24.04 LTS rather than Windows, alongside an Nvidia ConnectX-8 SuperNIC equipped with dual 400Gbps QSFP112 network interfaces.

Pricing for the ZGX Fury G1n reflects its niche infrastructure status: the baseline model starts at $132,000, rising to $133,000 for configurations with the RTX PRO 2000 and $136,000 when outfitted with the RTX PRO 4000.

Sources

  1. TechRadar Pro

Company: HP

Written by

The Company Wire

Newsroom · San Francisco

Inside the companies building what’s next. Reporting on startups, technology, funding and the people shaping them.