HP intros monstrous workstation PC with 496 GB LPDDR5X RAM and 252 GB VRAM

HP, in a collaboration with Red Hat, has announced a new enterprise-grade workstation PC dubbed the ZGX Fury AI Station. As the name suggests, the workstation is primarily targeted at AI-based workloads such as training, inferencing, and more.

Like the MSI WS300 we covered earlier, the ZGX Fury will be powered by Nvidia's GB300 Grace Blackwell Ultra chip. The chip sports a powerful 72-core ARM CPU based on Neoverse V2 cores, paired with an incredible 496 GB of LPDDR5X RAM.

The GPU side of things is equally impressive, with the system featuring a Blackwell Ultra GPU with 252 GB of dedicated HBM3e memory with an equally impressive memory bandwidth of 7.1 TB/s. Nvidia claims that the GPU can tap into the entire 748 GB memory pool and is good enough for 20 petaflops of FP4 compute (sparse).

Customers can also opt for discrete Nvidia RTX Pro GPUs, courtesy of the ZGX Fury workstation PC's PCIe Gen 5 x16 slot. Two more PCIe Gen 5 x16 slots are present, both with x8 speeds. A total of four M.2 storage slots are also present.

Moreover, folks who would like to pair multiple of these systems can utilize the Nvidia ConnectX-8 NIC to take advantage of up to 800 Gbps (dual QSFP112) of networking speeds. The rest of the I/O on offer has been detailed below:

• Six USB 3.2 Gen 2 Type-A
• Dual USB 3.2 Gen 2 Type-C
• 3.5 mm audio jack
• Dedicated microphone and audio outputs
• 1G Ethernet, 10G Ethernet

HP has not publicly disclosed to us how expensive the ZGX Fury workstation PC will be, but MSI's similarly equipped workstations were listed at around $100,000 recently, which indicates a similar price tag for the HP might be possible.

For normal users and other AI enthusiasts without $100,000 to spare, the AMD Strix Halo-powered GMKtec Evo X2 with 128 GB RAM (currently $3,649 on Amazon) might be a decent starting point for local LLMs.