--- title: "HP Launches ZGX Fury G1n AI Station Workstation: 748 GB Memory and 20 PFLOPS for Local AI" description: "HP has started sales of the ZGX Fury G1n AI Station workstation in the US: an Nvidia GB300-based system with 748 GB of memory and up to 20 PFLOPS of AI performance, with prices starting from $132,000." date: 2026-11-11T21:46:27.185Z lang: en url: https://xab.info/en/posts/hp-zgx-fury-g1n-ai-station-nvidia-gb300 tags: [hp, nvidia, ai-workstation, gb300, blackwell, local-ai, inference] author: "Alexander Buran" publisher: "XAB.info" --- # HP Launches ZGX Fury G1n AI Station Workstation: 748 GB Memory and 20 PFLOPS for Local AI ![Black tower chassis of the HP ZGX Fury workstation featuring a signature mesh front and logo](https://xab.info/media/2026/11/11/hp-zgx-fury-g1n-ai-station-nvidia-gb300-748-gbayt/hp-zgx-fury-g1n-ai-station-nvidia-gb300-748-gbayt-1.webp) ## Key Points - HP has started US sales of the ZGX Fury G1n AI Station workstation powered by the Nvidia GB300 Grace Blackwell Ultra Superchip. - The system is equipped with 496 GB LPDDR5X and 252 GB HBM3e, totaling 748 GB of memory, delivering up to 20 PFLOPS in AI tasks (FP4). - Available in three configurations: base with integrated GPU, and two with RTX PRO Blackwell (2000 at $133,000 and 4000 Blackwell SFF at $136,000), starting from $132,000. - Tower chassis with liquid cooling, 5U rack mountable, up to 32 TB storage, three PCIe 5.0 slots, two QSFP112 ports (400 Gbps), 1600W PSU, Ubuntu 24.04 LTS. - Designed for local inference of models up to 1 trillion parameters (FP4) and requires a 200–240V power grid. HP has started sales of the ZGX Fury G1n AI Station workstation in the US market. The device is positioned not as a desktop PC for mass consumers, but as a specialized system for organizations that require local execution of artificial intelligence models. At the heart of the machine is the Nvidia GB300 Grace Blackwell Ultra Superchip platform, and the total available memory reaches 748 GB — a figure that makes the system one of the highest-capacity in its class for large language model inference. ### Nvidia GB300 Platform and Record Memory Capacity The heart of the workstation is the Nvidia GB300 superchip, which combines a 72-core Neoverse V2 Grace Arm processor and a Blackwell Ultra graphics accelerator. The 496 GB of LPDDR5X RAM provides a bandwidth of up to 396 GB/s, while the 252 GB of HBM3e video memory delivers up to 7.1 TB/s. In total, the system features 748 GB of memory, allowing it to perform inference on AI models up to one trillion parameters in size with FP4 quantization. AI performance in FP4 format, according to the manufacturer, reaches 20 PFLOPS. ### Three Configurations and Price Range HP is releasing the ZGX Fury in three configurations. The base model is limited to the integrated Blackwell Ultra GPU, while the two higher-end versions are supplemented with discrete Nvidia RTX PRO Blackwell series graphics cards. The configuration with the RTX PRO 2000 and 16 GB of GDDR7 memory is priced at $133,000, while the RTX PRO 4000 Blackwell SFF variant equipped with 24 GB of memory costs $136,000. The starting price of the lineup, according to the announcement, begins at $132,000. ### Chassis, Cooling, and Interfaces The workstation is built in a tower chassis with liquid cooling; if necessary, it can be installed in a server rack, where it takes up 5U. It supports storage drives with a total capacity of up to 32 TB, and the board features three PCIe 5.0 expansion slots and two QSFP112 ports with a bandwidth of 400 Gbps each. A 1600W power supply unit is used for power, and the software package includes the Ubuntu 24.04 LTS operating system and a set of Nvidia tools. ### Target Audience and Power Requirements The ZGX Fury G1n AI Station is targeted at companies and research teams for whom local, rather than cloud-based, AI execution is critical. An important practical limitation is the power grid requirement: the system requires a 200–240V power line, which is uncommon in the US and may require the buyer to prepare separate infrastructure. Thus, HP offers a niche yet powerful solution for local inference of ultra-large models, where memory capacity and bandwidth become key performance factors. ## Fact-Check & Sources - [ПК по цене однушки в Москве: HP выпустила рабочую станцию ZGX Fury с 748 Гбайт памяти по цене от $132 000](https://3dnews.ru/1151175) - Primary source: GB300 platform, 496 GB LPDDR5X + 252 GB HBM3e = 748 GB, 20 PFLOPS (FP4), three configurations priced at $133,000 and $136,000, 5U chassis, 32 TB, PCIe 5.0, QSFP112 400 Gbps, 1600W PSU, Ubuntu 24.04 LTS, 200–240V power. - [ПК по цене однушки в Москве: HP выпустила рабочую станцию ZGX Fury с 748 Гбайт памяти по цене от $132 000](https://3dnews.ru/1151175/pk-po-tsene-odnushki-v-moskve-hp-vipustila-rabochuyu-stantsiyu-zgx-fury-s-748-gbayt-pamyati-po-tsene-ot-132000) - Duplicates primary source; confirms US sales launch, Nvidia GB300 platform, and 496 GB LPDDR5X. ## FAQ ### What platform is the HP ZGX Fury G1n AI Station built on? The system is built on the Nvidia GB300 Grace Blackwell Ultra Superchip platform, which includes a 72-core Neoverse V2 Grace Arm processor and a Blackwell Ultra graphics accelerator. ### What is the total memory capacity of the workstation? A total of 748 GB: 496 GB of LPDDR5X RAM and 252 GB of HBM3e video memory. ### What AI models can the system run locally? The system is capable of performing inference on artificial intelligence models up to one trillion parameters in size with FP4 quantization. ### How much does the HP ZGX Fury G1n AI Station cost? The starting price of the lineup is from $132,000. The configuration with RTX PRO 2000 (16 GB GDDR7) costs $133,000, and with RTX PRO 4000 Blackwell SFF (24 GB) it is $136,000. ### What are the power grid requirements for the system? Operation requires an electrical line with a voltage of 200–240V, which is uncommon in the US.