AMD Powers OpenAI’s GPT-OSS 20B & 120B AI Models on Ryzen & Radeon: Ryzen AI MAX+ 395 is the Sole AI Chip to Support 120B Model with Extraordinary 128 GB Memory Pool

OpenAI has unveiled its latest AI models, the GPT-OSS 20B and GPT-OSS 120B, and AMD is at the forefront of this innovation. The company has announced complete support for these models through its Ryzen AI MAX and Radeon GPUs, enabling users to leverage advanced capabilities and performance enhancements.

AMD’s Ryzen AI MAX+ 395 APU: A Game Changer for OpenAI’s GPT-OSS 120B

With the introduction of OpenAI’s new AI models, AMD has positioned its Ryzen AI CPUs and Radeon GPUs as the go-to hardware for optimal performance. Specifically, the Ryzen AI MAX+ 395 APU is highlighted as the exclusive chip that can execute the GPT-OSS 120B model natively, while also providing Day-0 support, allowing users to experience the models via LM Studio immediately.

AMD Ryzen AI Max+ leveraging OpenAI's GPT-OSS 120B with MCP support for enhanced processing.

What exactly are these new models? The GPT-OSS series comprises open-weight models capable of comprehensive reasoning and agentic tasks. While many AI chips and PCs can manage the 20B version, the more demanding 120B model necessitates significant hardware resources. This is where AMD’s Ryzen AI MAX and Strix Halo architectures shine, featuring up to 128 GB of memory that caters specifically to such advanced AI functionality.

Exploring advanced AI capabilities with AMD's systems.

The GGML converted MXFP4 weights require approximately 61 GB of VRAM, fitting seamlessly within the 96 GB dedicated graphics memory of the AMD Ryzen AI MAX+ 395 processor. Users need to ensure their driver version is AMD Software: Adrenalin Edition 25.8.1 WHQL or higher to utilize this feature efficiently.

With capabilities reaching speeds of 30 tokens per second, AMD users can access a powerful, datacenter-grade model. This performance is further enhanced by the bandwidth of the Ryzen AI MAX+ platform in conjunction with the innovative mixture-of-experts architecture found in the GPT-OSS 120B. Thanks to its extensive memory, users can also benefit from Model Context Protocol (MCP) implementations with this model. Notably, those with AMD Ryzen AI 300 series processors can fully leverage the smaller 20B model.

For optimal performance with the GPT-OSS 20B model, users are encouraged to utilize the AMD Radeon 9070 XT 16GB graphics card. This configuration not only provides exceptional speeds but also demonstrates impressive time-to-first-token (TTFT) advantages, particularly when working with Model Context Protocol (MCP) implementations in compute-heavy scenarios.

How to Experience OpenAI’s GPT-OSS 120B and 20B Models on AMD Hardware

Download and install the latest AMD Software: Adrenalin Edition 25.8.1 WHQL drivers or above. Be mindful that older drivers may compromise performance and compatibility.
For users with an AMD Ryzen AI-enabled machine, navigate to your Desktop and select AMD Software: Adrenalin Edition > Performance Tab > Tuning Tab> Variable Graphics Memory.Set VGM as per the specifications outlined in the accompanying table. If you are using an AMD Radeon graphics card, you can skip this step.
Install LM Studio on your system.
When prompted, choose to skip the onboarding process.
Search for “gpt-oss” in the application. You should find an option prefixed by “LM Studio community.”Select either the 20B or 120B variant based on your hardware compatibility.
Access the chat tab within LM Studio.
Use the drop-down menu to select the desired OpenAI model, ensuring you check “Manually load parameters.”
Adjust the “GPU Offload” slider to the maximum setting and enable the remember settings option.
Click the load button. Note that while loading the 120B model, it may take time, and the loading bar might appear to stall due to the size of the model.
Begin engaging with the model through prompts!

AMD Product Support Matrix detailing compatibility with OpenAI models.

AMD has also released a support list for OpenAI’s GPT-OSS models. Its Ryzen AI MAX+ 395 stands out as the sole chip capable of running the 120B model. In contrast, other options like the Radeon RX 9000, Radeon AI PRO R9000, and Radeon RX 7000 GPUs, all equipped with at least 16 GB of memory, can handle the GPT-OSS 20B models with ease.

Source & Images