AMD's Threadripper Halo is a local-AI workstation for researchers with deep pockets
AI workstation promises to put up to 576 GB of HBM3e and 16 TB/s of memory bandwidth on your desk
AMD has unveiled a workstation called Threadripper Halo, designed for researchers working on local AI applications. Unveiled at the IFA 2026 event, the system will debut next year and marks AMD’s first foray into offering Instinct accelerators in a workstation form factor. Similar to a previously reviewed AI Halo, Threadripper Halo isn’t bringing anything new to the table.
It’s powered by the 96-core Threadripper PRO 9995WX CPU, which debuted in 2025. Backed by up to 2 TB of DDR5 memory and 2.6 TB of combined system memory, the box is marketed as a local AI workstation for researchers. The system can be configured with up to four PCIe-based Instinct GPUs, although the IFa showcase only had two installed.
Launched in May, the MI350P is essentially half of an MI350X, packed into a PCIe form factor. Packing 144 GB of HBM3e memory, each accelerator delivers 4 TB/s of memory bandwidth and up to 4.6 petaFLOPS of FP4 compute. The Threadripper Halo promises up to 576 GB of HBM3e memory and 16 TB/s of memory bandwidth, enabling it to run models with over a trillion parameters in size at four-bit precision.
This includes running the largest open weights models, such as Moonshot.AI’s 2.8 trillion-parameter Kimi K3. However, the system won’t come cheap and may face availability restrictions in certain markets. Pricing hasn’t been disclosed, but it’s expected to range between $100,000 and $150,000. With its quad MI350P configuration, the system pushes the limits of standard North American power outlets, unless AMD underclocks the cards or mandates a 20 amp circuit.
Positioned against Nvidia’s high-end AI workstation, the DGX Station, Threadripper Halo offers up to 3.4x the total system memory and more than twice the memory bandwidth.
Written by urgent.news from The Register's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.