AMD Creates a Threadripper Desktop PC for AI Models

At IFA in Berlin this week, AMD introduced a new powerhouse desktop PC line, Threadripper Halo Station, designed to run AI models locally.

"This is the most powerful workstation in the world,"  says  AMD SVP and General Manager Jack Huynh. "Designed and engineered for a complete new era of computing, capable of running AI models that are a trillion parameters ."

The product is meant for businesses and software developers who prefer running AI in their office rather than in the cloud. The liquid-cooled desktop PC features a 96-core Threadripper Pro 9995WX CPU and two enterprise-grade MI350P GPUs . Buyers can even expand it to four GPUs. "This is about as close as you can get to a personal supercomputer," Huynh adds. 

Pricing and launch date were not announced. But the product will likely be extremely expensive, given that it can feature up to 2TB of system memory and up to 576GB of high-bandwidth HBM3E memory if equipped with four GPUs. 

Threadripper Halo Station appears poised to compete with Nvidia's DGX Station tower PCs, which support AI models spanning up to 1 trillion parameters, for at least $100,000 .

Project Zenith Coming to Ryzen AI Halo PCs

AMD's other notable announcement is Microsoft's Project Zenith, a special flavor of Windows 11 designed for software developers and coders, which we first saw at Build . Zenith will arrive on AMD's Ryzen AI Halo PCs, which launched in July as mini PCs for $3,999. Micro Center has been selling units with Windows 11 Pro or Linux.  

Project Zenith stands out for offering a suite of preinstalled developer tools and focusing on a "distraction-free Windows experience." Microsoft also created the OS for devices with 64GB or more of unified memory, enabling them to run large AI models locally.

We don't have pricing or a launch date, but Microsoft says , "Project Zenith is just the beginning," while hinting that it's working with a range of PC makers on more developer-friendly systems.

PCMag and Yahoo may earn commission from links in this article.