AMD’s Ryzen AI MAX+ 395 is the only AI chip capable of running OpenAI’s GPT-OSS 120B model natively. With a maximum memory pool of 128 GB, these chips support both the 20B and 120B models.

Login to Continue Learning

Yesterday, OpenAI released two new AI models: GPT-OSS 20B and GPT-OSS 120B. AMD’s Ryzen AI CPUs and Radeon GPUs offer Day-0 support for these models and are available to try out via LM Studio.

GPT-OSS models are designed to handle complex reasoning and agentic capabilities. While most AI PCs can manage the 20B model, the 120B model requires more resources. AMD’s Strix Halo or Ryzen AI MAX chips, with a maximum memory pool of 128 GB, are capable of handling such models.

The GGML converted MXFP4 weights for the GPT-OSS 120B model require around 61GB of VRAM and fit comfortably into the 96GB dedicated graphics memory of the AMD Ryzen AI Max+ 395 processor. A driver version equal to or higher than AMD Software: Adrenalin Edition 25.8.1 WHQL is required for this functionality.

With up to 30 tokens per second, the performance is very usable thanks to the bandwidth of the Ryzen AI Max+ platform and the mixture-of-experts architecture of the OpenAI GPT-OSS 120B model. The Ryzen AI Max+ 395 (128GB) also supports Model Context Protocol (MCP) implementations with this model.

Users with AMD Ryzen AI 300 series processors can utilize the smaller 20B model from OpenAI. For lightning-fast performance, users can use an AMD Radeon 9070 XT 16GB graphics card in a desktop system for the GPT-OSS 20B model. This setup offers impressive TTFT advantage and responsive performance with MCP implementations.

To experience these models on AMD Ryzen AI processors and Radeon graphics cards:

1. Install AMD Software: Adrenalin Edition 25.8.1 WHQL drivers or higher.
2. If using an AMD Ryzen AI machine, adjust the Variable Graphics Memory as specified.
3. Download and install LM Studio.
4. Skip onboarding.
5. Go to the discover tab (magnifying glass).
6. Search for “gpt-oss” and select either 20B or 120B.
7. Go to the chat tab, manually load parameters, and move GPU Offload to MAX.
8. Click load; if using the 120B model, it may take time due to large read speeds.

AMD has provided a product support matrix for OpenAI models, with the Ryzen AI MAX+ 395 being the only chip capable of handling the 120B model. Other compatible Radeon GPUs include those with at least 16 GB memory.

📚 Reading Comprehension Quiz

10 Points

What is required to handle the GPT-OSS 120B model according to the content?

A. Most AI PCs B. AMD's Strix Halo or Ryzen AI MAX chips with a maximum memory pool of 128 GB C. AMD Ryzen AI 300 series processors D. An AMD Radeon 9070 XT 16GB graphics card

Please login or register to take the quiz and earn points!