AMD Unveils 'Ryzen AI Halo,' a Powerful AI Mini-PC with 128GB RAM for Developers, Directly Challenging NVIDIA
AMD, in collaboration with retailer Micro Center, launched 'Ryzen AI Halo,' a compact computer for AI developers packed with powerful specifications, prominently featuring up to 128GB of unified memory to enable the execution of large language models on a single device.
AMD has launched the Ryzen AI Halo, a mini-PC device specifically designed for on-device AI development (local AI). It is exclusively sold at Micro Center stores in the U.S.
The core of the machine is the new AMD Ryzen AI Max+ 395 processor, which features a 16-core Zen 5 CPU, 40 RDNA 3.5 graphics units, and crucially, an XDNA 2 architecture NPU (Neural Processing Unit) delivering up to 50 TOPS (Trillion Operations Per Second) of AI processing power.
The most distinctive feature that sets the Ryzen AI Halo apart is its massive 128GB of LPDDR5x unified memory, boasting a maximum bandwidth of 256GB/s. AMD states this is sufficient to run AI models with up to 200 billion parameters directly in the device's memory, enabling developers to work with large models without relying on the cloud.
The device is compact, measuring only 150 x 150 x 45.4 mm and weighing approximately 1.2 kg. It comes equipped with a full suite of workstation-grade connectivity ports, including 10Gbps Ethernet, two USB-C 4.0 ports, one USB-C 3.2 port, and HDMI 2.1. It is available in both Linux and Windows operating system versions.
The Ryzen AI Halo is priced at $3,999 USD, positioned as a direct competitor to the NVIDIA DGX Spark, which retails for $4,700 USD. Micro Center, the distributor, claims performance figures showing the ability to run the GPT-OSS-120b model at 45 tokens per second, although these figures have not yet been independently verified.
The Ryzen AI Halo presents an interesting new option for AI developers and small businesses who need tools to run large language models on-premises (local AI) without relying on the cloud, and at a more accessible price point than direct competitors like NVIDIA.