AMD Ryzen AI MAX+ Processors Unleash 96 GB Memory for Consumer Graphics, Enabling Local Execution of 128B-Parameter LLMs on PCs

AMD has significantly advanced edge AI capabilities for personal computers with its latest driver update, enabling support for large language models (LLMs) to users.

AMD Empowers Consumer AI Innovations with Strix Halo Platform

As a prominent player in the industry, AMD has made remarkable strides in integrating high-performance AI computing within consumer systems through its innovative XDNA engines. The company’s Accelerated Processing Units (APUs), particularly those in the Strix Point and Strix Halo lineups, demonstrate exceptional AI capabilities. The latest release of the Adrenalin Edition 25.8.1 driver introduces support for LLM models that can reach up to 128 billion parameters, paving the way for consumer devices to harness advanced AI models, including Meta’s cutting-edge Llama 4 Scout.

AMD Ryzen AI: Leading Windows LLM Support, 4-bit to 16-bit Quantization, Thin and Light Deployment

One of the standout features of AMD’s technology is its Variable Graphics Memory (VGM), which allows consumers to allocate up to 96 GB of graphics memory to the integrated GPU (iGPU).This capability enables the execution of large-scale AI models locally. For instance, the Llama model operates with a mixture of experts (MoE) design and only engages 17 billion parameters during runtime. Consequently, users can achieve impressive transactions per second (TPS), empowering the LLM to function efficiently as a sophisticated AI assistant.

AMD AI Max+ runs up to 128B parameters; memory usage per model comparison chart.

AMD’s advancements extend beyond sheer computational power; they have also made significant improvements in context size. Traditionally, consumer processors support a context length of approximately 4096 tokens. However, AMD has achieved an impressive context length of 256, 000 tokens, greatly enhancing user control over workflows and consistently ensuring robust performance. These advancements illustrate the fervent progression of “AI compute”technology in personal computers, marking a notable shift in the industry.

Currently, the Strix Halo platform is available on a limited number of devices, and accessibility remains a challenge for many consumers, with some systems priced above $2, 000. Despite the elevated cost, AMD’s commitment to democratizing AI power is evident, even though acquiring these cutting-edge devices requires a significant financial investment.

Source & Images