Quickly scale and streamline AI development, inference, and agentic AI workflows with groundbreaking on-prem compute and a flexible software stack. The HP ZGX Fury AI station hosts large-scale models up to 1 trillion parameters1 locally, helping keep data secure and costs predictable. Liquid cooling and optimized airflow help support consistent performance across demanding AI workflows. Airflow is drawn from the front and exhausted out the back, so the system can be racked without restricting ventilation.
Maintain greater control over data, infrastructure, and costs with powerful on-prem AI.
Enable multiple users and concurrent AI workloads from a single AI system.
Purpose-built AI infrastructure designed specifically for modern AI.
Power AI development and production-grade inference on-prem and leave token fees behind.
NVIDIA GB300 Grace Blackwell Ultra Superchip brings incredible AI performance on-prem, with up to 20 petaFLOPS FP4 and 748 GB unified memory.
ZGX Fury G1n comes preinstalled with Ubuntu 24.04 LTS with NVIDIA AI Developer Tools - drivers, CUDA, and container tools ready to go. Start training and deploying models the moment you power on.
Effortlessly develop and fine-tune up to 100 billion class or inference up to trillion parameter class AI models with 748 GB of large, coherent memory while you offload video to an optional NVIDIA RTX PRO GPU.
Choose between 2 or 4 TB of NVMe M.2 self-encrypted storage3 to work efficiently with large files and keep more of your data local and secure.
HP Z Toolkit4 and HP Z Runtime - purpose-built tools for faster setup of scalable AI workflows. Discover remote from any device to a Z workstation. Pull and serve with one command on ZGX machines.
HP ZGX Fury is powered by Red Hat AI Enterprise for enhanced AI and orchestration capabilities to run production inference closer to users, applications, machines and data.
Run inference for multiple users simultaneously or give developer teams a shared platform to experiment and fine-tune in parallel on a single HP ZGX Fury. Link two devices together to further scale model capacity and performance. Its rack-mountable, office-deployable design plugs into standard enterprise environments
| 1. | Power Button | 8. | Dust Filter |
| 2. | Power LED (Blue) | ||
| 3. | Activity LED (White) | ||
| 4. | 2x USB 3.2 Gen 2.1 (5Gbps) Type-A ports | ||
| 5. | Microphone-In | ||
| 6. | Headphone-Out | ||
| 7. | 2x USB 3.2 Gen 2 (10 Gbps) Type-C ports |
| 1. | 20 AC Inlet | 2. | 2x QSFP112 (400Gbps) ports |
| 3. | 4x USB 3.2 Gen 2 (10 Gbps) ports | 4. | Line-out, Line-in, Universal audio jack ports |
| 5. | Micro-USB to BMC | 6. | Mini-DP to BMC |
| 7. | Power Switch | 8. | Hybrid Mode button/td> |
| 9. | RJ-45 (1 Gbps) to BMC | 10. | RJ-45 (10 Gbps) |
| 11. | Kensington Lock Slot |
| Operating Systems | Ubuntu 24.04 LTS with NVIDIA AI Developer Tools |
| Processor family | NVIDIA Grace Blackwell Ultra Desktop Superchip |
| Available Processors | NVIDIA GB300 Grace Blackwell Ultra Desktop (NVIDIA® Grace-72 Core Neoverse V2 CPU) with NVIDIA® Blackwell Ultra GPU (252 GB HBM3e) |
| Form factor | Liquid cooled rackable tower |
| Maximum memory | 496 GB LPDDR5X 6400 MT/s SOCAMM |
| Internal storage | 2 TB HP Z Turbo Drive PCIe M.2 SSD ; 4 TB HP Z Turbo Drive PCIe M.2 SSD ; 2 TB Z Turbo PCIe NVMe M.2 SED NMIC SSD |
| Additional storage | Four M.2 storage slots: Two PCIe Gen5 x4 M.2 devices for OS (RAID 1 mirrored) and two PCIe Gen6 x4 for data (RAID 0 striped). Drive types and capacities vary by configuration. |
| Available Graphics | Integrated: NVIDIA Blackwell Ultra GPU ; Discrete: NVIDIA RTX PRO 4000 Blackwell Generation (24 GB GDDR7 dedicated); NVIDIA RTX PRO 2000 Blackwell Generation (16 GB GDDR7 dedicated) |
| Expansion slots | 3 PCIe Slots: 1 PCIe Gen5 x16; 2 PCIe Gen5 x16 (x8 electrical) |
| Communications | NVIDIA ConnectX-8, 2x 400Gb QSFP112 ports (supports optical and DAC connectivity) |
| Software | HP Z Runtime; NVIDIA AI Developer Tools; NVIDIA approved partner-developed BIOS; NVIDIA approved partner-developed BMC Firmware |
| Management features | Out of band with NVIDIA integrated BMC; Redfish API support |
| Power | 1600 W Titanium |
| Weight | Starting at 41 kg |