HP has started selling the ZGX Fury G1n AI Station in the United States. The system is built on the Nvidia GB300 Grace Blackwell Ultra Superchip platform and is equipped with 496 GB of LPDDR5X RAM. There are also 252 GB of HBM3e video memory, that is, the total capacity is 748 GB.
Image source: HP
The Nvidia GB300 platform includes a 72-core Neoverse V2 Grace Arm CPU and Blackwell Ultra graphics. LPDDR5X-6400 memory offers up to 396 GB/s bandwidth, and HBM3e memory offers up to 7.1 TB/s. System performance in AI tasks (FP4 format) reaches 20 Pflops.
HP is releasing three configurations of the ZGX Fury. The base model is limited to an integrated Blackwell Ultra GPU, while the other two are complemented by Nvidia RTX PRO Blackwell series graphics cards. There is a model with RTX PRO 2000 and 16 GB of GDDR7 memory for $133,000, a configuration with RTX PRO 4000 Blackwell SFF is available – it has 24 GB of memory with a price tag of $136,000. The system is capable of performing inference, that is, local launch of artificial intelligence models of up to 1 trillion parameters in size with FP4 quantization.
The workstation is made in a tower case with liquid cooling; if necessary, it can be installed in a server rack, and it will occupy 5U. Drives with a total capacity of up to 32 TB are supported, there are three PCIe 5.0 expansion slots and two QSFP112 ports with a throughput of 400 Gbps each. A 1600 W power supply is installed on board; software package includes Ubuntu 24.04 LTS OS and Nvidia toolset.
The HP ZGX Fury G1n AI Station is not aimed at everyday desktop users, but at organizations that require an AI model running locally. The system requires a 200-240V electrical line to operate, which is uncommon in the US.
If you notice an error, select it with the mouse and press CTRL+ENTER.
