Product features
CPU & GPU
- Processor: 1x NVIDIA® Grace ™ 72-core Neoverse V2 CPU (ARM-based)
- GPU: 1x NVIDIA® Blackwell Ultra GPU
- MIG: Multi-instance GPU for up to 7 MIGs
- optional: 1x dedicated NVIDIA® Blackwell GPU for graphics rendering, e.g. RTX Pro™ 2000 Blackwell
RAM
- total coherent memory: 748GB, linked via NVLink ®-C2C, at 900 GB/s interconnect bandwidth
- CPU memory: 496GB ECC LPDDR5X, 396 GB/s memory bandwidth (SOCAMM, 6400 MT/s)
- GPU memory: 252GB ECC HBM3e, 7.1 TB/s memory bandwidth
Networking
- Dual-port QSFP112 400GbE LAN via NVIDIA ConnectX-8 SuperNIC (= 800 Gbit/s bandwidth total)
- 1x RJ45 10GbE LAN port
- 1x RJ45 1GbE Dedicated BMC LAN port
Storage
- 4x M.2 PCIe 5.0 x4 NVMe (M-key 2280)
- populated with 4x 2TB or 4x 4TB SSDs - configurable
internal storage options:
- 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
PCIe Slots
- 1x PCIe Gen 5 x16 FHFL double-width, for GPUs
- 2x PCIe Gen 5 x8 (in x16) FHHL
- 1x M.2 PCIe 2.0 x1 slot (E-Key 2230))
Front IO Ports and Connectors
- 2x USB 3.2 Gen2 port (Type-C, 10 Gbit/s)
- 2x USB 3.2 Gen1 ports (Type-A, 5 Gbit/s)
- 1x unviersal audio jack
Rear IO Ports and Connectors
- 4x USB 3.2 Gen2 ports (Type-A, 10 Gbit/s)
- 2x QSFP ports (400Gb per port)
- 1x RJ45 port 10 Gbit/s Ethernet
- 1x BMC management port RJ45 1 Gbit/s Ethernet
- 1x mini-DP 2.0 (video port, 1920 x 1080 @ 60 Hz)
- 1x microUSB port (for BMC management)
- 1x microphone port
- 1x audio-in
- 1x audio line-out
OS & security
- OS: Ubuntu 24.04 LTS with NVIDIA AI Developer Tools
- pre-installed with essential CUDA-X™ libraries
- HP Z Runtime, compatible with HP ZGX Toolkit for Visual Studio Code client devices
-
Trusted Platform Module (TPM) 2.0
- UEFI version 2.9
- UEFI secure boot
-
NVIDIA approved partner-developed BIOS
-
NVIDIA approved partner-developed BMC
- Kensington Cable lock
IT manageability and enterprise features
- out-of-band telemetry via BMC
- Redfish API support
- NVIDIA Data Center GPU Manager
- visibility, monitoring, remote management capabilities
- hardware root of trust
- enterprise secure boot
Dimensions, weight, form factor
- Form factor: desktop tower, rack-mountable with accessory kit (included)
- height: 446 mm / 17.56''
- width: 221.5 mm / 8.72''
- depth: 667 mm / 26.26''
- packaging dimensions: 806.5 x 589 x 435 mm (31.75 x 23.19 x 17.125 inch)
- net weight: starting at 41kg, exeact weight depends upon the configuration
System cooling
- cooling system: closed-loop liquid cooling (suitable for office use)
- liquid cooling: Direct to Chip (D2C) Cold Plate
Power supply
- power supply: 1x 1600W Titanium power supply, 100 - 240V
Misc
- certification and compliance: TAA compliant
- rack rails for rack configuration option included
- operating temperature: TBD
- storage temperature: TBD
- operating relative humidity: TBD (non-condensing)
- storage relative humidity: TBD (non-condensing)
AI performance
- FP4 Tensor core: 20 PFLOPS / 15 PFLOPS without sparsity
- FP8 / FP6 Tensor core: 10 PFLOPS
- Int8 Tensor Core: 330 TOPS
- FP16/BF16 Tensor Core: 5 PFLOPS
- TF32 Tensor Core: 2.5 PFLOPS
- FP32: 80 TFLOPS
- FP64 / FP64 Tensor Core: 1.3 TFLOPS
Product description
The HP ZGX Fury G1n AI Station is an NVIDIA DGX Station based on the GB300 Grace™ Blackwell superchip. It can be used for running local AI applications for development and production purposes.
Example AI applications include LLM inference for frontier models of up to 1 Trillion parameters, scientific workloads (e.g. BioNeMo), finetuning 100 billion class models, hosting AI agents, and running RAGs for AI-enhanced document workflows. In particular it is well-suited to run MoE (mixture of experts) models.
Thanks to its MIG feature and its massive memory throughput it can be run headless to support several colleagues in a local workgroup.
Compared to server systems, or heavy multi-GPU systems, it runs cooler and quieter, and is uniquely suited for office use. Its server-grade management features, such as the integrated BMC, the Redfish API support, and enterprise secure boot allow to integrate it into a corporate IT management landscape.
Two ZGX Fury G1n AI Station DGX Stations can be joined using the Dual-port QSFP112 400GbE LAN to serve even bigger models, increase precision, or general model performance.
The HP ZGX Fury AI Station offers two distinct features from other DGX Station vendors:
- Included HP ZGX Toolkit / HP ZU Runtime CLI software - this allows you to link two HP DGX Stations effortlessly, and quickly pull and deploy Hugging Face models, MLflow tracking and Ollama testing
- Rack-mount rails for rack configuration are included in the package
A separately obtainable NVIDIA AI Enterprise software license provides a suite of software tools, libraries, and frameworks to fast-track your AI deployment on the HP ZGX Fury DGX Station GB300.
Optional GPUs
One of the following optional GPUs can be added into the system to add graphics output options, and for scientific simulation and rendering workloads:
- NVIDIA RTX PRO™ 4000 Blackwell Generation (24 GB DDR7 dedicated)
- NVIDIA RTX PRO™ 2000 Blackwell Generation (16 GB DDR7 dedicated)
Datasheets, manuals and additional downloads
- TBD
Official business partner
pi3g GmbH & Co. KG (operating buyzero.de) is an official business partner of HP, reseller of the HP ZGX Fury G1n AI Station and other HP AI Solutions.
Lead time & purchase
This product is available on pre-order with a current lead time of about 8-12 weeks. Contact us to purchase it.
Additional SKUs and codes
- TBD