HP has launched the ZGX Fury AI station, a system designed for local AI development, fine-tuning, and production-grade inference. The station is built around the NVIDIA GB300 Grace Blackwell Ultra Superchip, providing up to 20 petaFLOPS of FP4 compute and 748GB of unified memory. To allow the GPU to focus entirely on inference, HP offers an optional NVIDIA RTX PRO GPU for display output.

The hardware configuration includes a 72-core Grace CPU with 496GB of LPDDR5X memory, paired with a Blackwell Ultra GPU featuring 252GB of HBM3e memory. The system is designed to support large-scale models of up to 1 trillion parameters when using FP4 quantization. For storage, the unit provides options for 2TB or 4TB of self-encrypting NVMe M.2 drives in a RAID 0 configuration, alongside additional M.2 slots for the operating system in RAID 1.

HP is collaborating with Red Hat to offer an enterprise-grade AI platform by integrating Red Hat AI Factory with NVIDIA. The ZGX Fury is certified for Red Hat Enterprise Linux, aiming to reduce deployment risks and simplify workload orchestration for hybrid cloud environments. The platform includes specialized software such as HP Z Runtime for model management and HP Z Toolkit for experiment tracking and testing.

The ZGX Fury is available as a tower or a 5U rack-mountable system, intended for use in departments, factories, or branch offices as shared edge infrastructure.

Sources: