On the eve of Computex, AMD officially launched the Ryzen AI Max 400 series processors (codenamed Gorgon Halo), serving as successors to the Strix Halo (Ryzen AI Max 300). The new series retains the large-scale SoC architecture featuring a Zen 5 CPU, RDNA 3.5 GPU, and XDNA 2 NPU. Top-tier configurations include 16 cores with 32 threads, a CPU boost clock of up to 5.2GHz, 40 GPU compute units running at a maximum frequency of 3.0GHz, and an NPU capable of delivering up to 55 TOPS of computing power. The biggest upgrade this time is the expansion of unified memory capacity from a maximum of 128GB in the previous generation to 192GB, resulting in up to 160GB of available graphics memory. AMD claims this makes it the first processor on the x86 client platform able to run large language models with over 300 billion parameters locally. Currently, AMD has only announced PRO variant SKUs; whether non-PRO versions will follow remains unconfirmed.
Meanwhile, since the underlying memory bandwidth saw only a marginal increase (LPDDR5X speeds rose from 8000MT/s to roughly 8533MT/s), industry analysts point out that existing bottlenecks during the AI inference prefill phase will persist. The practical benefits of having 192GB of memory are primarily limited to scenarios involving ultra-large models where memory capacity is a major constraint. Systems equipped with Ryzen AI Max PRO 400 processors will be released by ASUS, HP, and Lenovo starting in Q3 2026. The only confirmed model at present is AMD’s own Ryzen AI Halo workstation, powered by the flagship Ryzen AI Max+ Pro 495; its starting price has yet to be disclosed. Additionally, a version featuring the previous-generation Max+ 395 is expected to become available for pre-order in June, priced from $3,999.