12:53Machine-translated from Azerbaijani by AIRead in Azerbaijani

HP Unveils $132,000 AI Workstation

HP has released the ZGX Fury G1n workstation for AI workloads that require high computing power. The model is powered by an Nvidia GB300 Grace Blackwell Ultra processor. Its starting price is $132,000 (approximately 224,400 AZN). The system has a total of 748 GB of memory. This amount of memory helps run large AI models directly on the device without connecting to a remote server.

Of the memory, 496 GB is LPDDR5X, while the remaining 252 GB uses the HBM3e standard. Together, the two types provide a total memory capacity of 748 GB. The LPDDR5X memory operates at 6,400 MT/s and is connected to a 72-core Arm Neoverse V2 Grace processor. The faster HBM3e memory is built into the Blackwell Ultra graphics processor. Depending on the optional graphics configuration selected, video memory can reach 276 GB.

According to HP, language models with up to 100 billion parameters can be fine-tuned locally on the workstation. Using a compression method called FP4, the system can also handle the response-generation process for models with 1 trillion parameters. However, actual speed will depend on the model selected, the amount of data and the task being performed.

A liquid cooling system has been installed to prevent the powerful hardware from overheating. The power supply is rated at 1,600 W and has Titanium certification. The workstation comes with the Ubuntu 24.04 LTS operating system installed. The Nvidia ConnectX-8 SuperNIC network module reaches speeds of 400 Gbps. The module has 2 QSFP112 connections, allowing multiple systems to be linked and operated together. The chassis can also be installed in a 5U equipment rack.

The standard configuration will sell for $132,000. The version with an RTX PRO 2000 graphics card will cost $133,000. This is approximately 226,100 AZN. The RTX PRO 4000 version will sell for $136,000. Its price in AZN is approximately 231,200. At these prices, the ZGX Fury G1n is not accessible to users looking for an ordinary desktop computer. The system is primarily intended for companies and research institutions that want to run large AI models on their own hardware.