Quote:AMD Ryzen AI MAX+ with 128GB now support Large Language Models up to 128 Billion parameter



AMD has just rolled out a major upgrade for its Ryzen AI Max+ platform, expanding support for massive LLMs on Windows. This builds on what they showed at CES 2025, where the Ryzen AI processors became the first to run Llama 70B locally using llama.cpp and LM Studio. With the upcoming Adrenalin Edition 25.8.1 WHQL driver, systems like the Ryzen AI Max+ 395 with 128GB RAM will now be able to handle models with up to 128 billion parameters in Vulkan llama.cpp. That’s thanks to AMD’s Variable Graphics Memory, which allows up to 96GB to be allocated for AI workloads.

This update also makes the Ryzen AI Max+ 395 the first Windows AI processor to run Meta’s Llama 4 Scout 109B (with 17B active) locally, complete with vision and MCP support. Since Llama Scout uses a mixture-of-experts setup, only part of the model is active at a time, which keeps performance usable, up to 15 tokens per second in testing. The platform supports a wide range of model sizes and quantization formats, including GGUF, making it very flexible for different use cases.

Continue Reading...