AMD driver update significantly improves AI processing power, enabling 109B's Llama 4 Scout to run locally
AMD's AI chip, the
AMD Ryzen™ AI Max+ Upgraded: Run up to 128 Billion parameter LLMs on Windows with LM Studio
https://www.amd.com/en/blogs/2025/amd-ryzen-ai-max-upgraded-run-up-to-128-billion-parameter-llms-lm-studio.html
On July 29, 2025, AMD released an update to its 'AMD Variable Graphics Memory' that enables support for up to 128 billion parameters in Vulkan llama.cpp on Windows. This update will be included in the upcoming 'Adrenalin Edition 25.8.1 WHQL' driver. The update will enable memory-intensive AI workloads and allow the Ryzen AI MAX+ 395 (128GB) on Windows to take full advantage of the 96GB VGM.
With this update, the AMD Ryzen AI Max+ 395 (128GB) becomes 'the world's first Windows AI PC processor capable of running Meta's Llama 4 Scout 109B with full vision and MCP support.'
Meta Llama 4 Scout is an MoE model that contains 16 expert models optimized for each task. While the total number of parameters is 109 billion, only 17B (17 billion) of the parameters are active at any one time. However, because all 109 billion parameters must be stored in memory, AMD explains that memory usage is the same as a 109 billion parameter model.
Users can enjoy a maximum speed of 15 tokens per second. See the video below for an example of how it works.
AMD Ryzen™ AI Max+ 395 Upgraded: Meta Llama 4 Scout Demo - YouTube
The AMD Ryzen AI Max+ 395 can also be installed in a laptop, making it easy to try out a high-performance model on the go.
Another feature is that Llama 4 Scout can be run with a context length of 256,000, which allows for a large number of tokens to be stored in the context, enabling powerful 'agent-based' workflows.
The demo below shows how to retrieve and summarize AMD's quarterly financial report. Among other things, the AI needs to keep 19,642 tokens in the context. The default context window limits the context length to 4096, which could cause the process to fail. However, with this update, the process can now be completed without any issues.
AMD Ryzen™ AI Max+ 395 Upgraded: ARXIV MCP Demo - YouTube
The new update can be tried out by downloading the preview driver and LM Studio. AMD stated, 'The AMD Ryzen AI Max+ 395 further strengthens the industry-leading Windows platform advantage in thin and light systems.'
Related Posts:
