HP has introduced the ZGX Fury G1n, a powerful AI workstation equipped with NVIDIA's GB300 Grace Blackwell Ultra superchip and a combined 748GB of memory.
The system starts at $132,000 and is designed for organizations developing large artificial intelligence models, running local inference workloads and handling demanding research applications.
Despite using hardware typically associated with data centers, the ZGX Fury comes in a desktop tower that can also be installed in a standard server rack.
Its configuration includes a 72 core NVIDIA Grace CPU, a Blackwell Ultra GPU with 252GB of HBM3e memory and 496GB of LPDDR5X system memory.
HP is positioning the workstation as an alternative to cloud computing infrastructure for businesses that require substantial computing resources within their own facilities.
NVIDIA GB300 Combines 72 CPU Cores With 252GB of HBM3e Memory
The ZGX Fury is built around NVIDIA's GB300 Grace Blackwell Ultra platform.
Its Grace processor features 72 Arm Neoverse V2 CPU cores paired with 496GB of LPDDR5X memory operating at 6400 MT/s.
The Blackwell Ultra GPU provides an additional 252GB of high bandwidth HBM3e memory for demanding AI processing tasks.
Together, these components offer 748GB of combined CPU and GPU memory capacity.
However, the two memory pools serve different purposes and should not be confused with a single conventional RAM configuration.
HP also offers configurations with additional professional graphics cards, including NVIDIA RTX PRO 2000 and RTX PRO 4000 models.
Depending on the configuration, the system can provide up to 276GB of graphics memory.
HP Targets Local AI Models With Up to One Trillion Parameters
The large memory configuration makes the ZGX Fury suitable for running AI models that would be difficult to accommodate on conventional desktop workstations.
HP claims the system can support local development and fine tuning of language models containing up to 100 billion parameters.
It also advertises inference support for models with as many as one trillion parameters when using FP4 quantization.
Actual performance and memory requirements will depend on the model architecture, context length and workload.
The workstation could appeal to organizations working with sensitive information because local AI processing can reduce the need to transfer data to external cloud services.
However, the high initial purchase price means the financial benefits will depend on how frequently the system is used and the cost of alternative computing resources.
HP ZGX Fury Specifications and Pricing
| Component | Specification |
|---|---|
| Platform | NVIDIA GB300 Grace Blackwell Ultra |
| CPU | 72 core Arm Neoverse V2 Grace |
| System memory | 496GB LPDDR5X 6400 |
| GPU | NVIDIA Blackwell Ultra |
| GPU memory | 252GB HBM3e |
| Combined memory | 748GB |
| Networking | Dual 400Gbps QSFP112 |
| Network controller | NVIDIA ConnectX 8 SuperNIC |
| Cooling | Liquid cooling |
| Power supply | 1600W Titanium certified |
| Operating system | Ubuntu 24.04 LTS |
| Form factor | Desktop tower, 5U rack compatible |
| Starting price | $132,000 |
The dual 400Gbps networking connections are particularly important for organizations planning to connect multiple workstations.
They allow systems to exchange large amounts of data through high speed networking infrastructure, supporting more demanding distributed AI workloads.
HP has also included liquid cooling to manage the thermal requirements of the GB300 platform inside a relatively compact chassis.
HP Offers RTX PRO Configurations Starting at $133,000
The ZGX Fury is available in several configurations, depending on the additional graphics hardware required.
| Configuration | Starting price |
|---|---|
| HP ZGX Fury G1n | $132,000 |
| ZGX Fury with NVIDIA RTX PRO 2000 | $133,000 |
| ZGX Fury with NVIDIA RTX PRO 4000 | $136,000 |
The pricing places the workstation firmly in the enterprise and research markets rather than the conventional desktop PC segment.
For organizations that regularly run large AI models, the ZGX Fury offers substantial local computing and memory capacity without requiring a traditional data center installation.
Its combination of NVIDIA's GB300 architecture, extensive memory and high speed networking also makes it suitable for building smaller AI computing clusters.
With a starting price of $132,000, however, the system is intended for specialized workloads where its hardware capabilities can justify the investment.
Discussion (0)
Be the first to comment.