AMD has opened a new horizon for AI computing by enabling data center-grade AI models to run on consumer PCs.

AMD announced a hardware platform that immediately supports GPT-OSS 20B and 120B, the first openweight language models released by OpenAI on the 5th.

This marks the beginning of an era in which data center-grade AI models can be run on consumer PCs.

GPT-OSS 120B is a very large model with 510 million of the 11.68 billion parameters activated.

GPT-OSS 20B is evaluated as a lightweight, high-performance model with 360 million of its 2.09 billion parameters activated.

Both models are groundbreaking in that they have advanced inference capabilities and can run in a local environment.

AMD unveiled the Ryzen™ AI Max+ 395, the first consumer AI PC processor capable of running the GPT-OSS 120B model locally.

This processor supports up to 128 GB of memory and can run models with up to 128 B parameters based on llama.cpp.

In particular, it can easily handle MXFP4 models requiring 61GB of VRAM by utilizing 96GB of dedicated graphics memory.
/> In a desktop environment, the GPT-OSS 20B model can run at ultra-fast speeds through the AMD Radeon™ 9070 XT 16GB graphics card.

In particular, when implementing MCP (Model Context Protocol), TTFT (Time to First Token) performance is excellent, providing fast response even in computing-intensive situations.

Users can download the GPT-OSS model via LM Studio and install the AMD Adrenalin Edition 25.8.1 WHQL or later driver.

You can adjust the GPU offload settings to the maximum and manually load the parameters to get immediate prompt input.

This announcement demonstrates that AMD is the only vendor that encompasses AI workloads from cloud to client. By enabling data center-grade models to run on thin and light laptops, the democratization of AI is expected to accelerate.