Ryzen AI Software 1.7 brings some hearty performance improvements when it comes to its CNN/Transformer compiler with much better performance and quicker compile times. There is also support for various new large language models (LLMs) on Ryzen AI NPUs like Qwen-2.5-14b-Instruct, Qwen-3-14b-Instruct, Phi-4-mini-instruct. Plus in preview form is Sparse-LLM support for the GPT-OSS-20b NPU model and VLM Gemma-3-4b-it. There is also now long context support for hybrid execution LLM models.
On the Stable Diffusion side there is new model support for SD3.5-Turbo with 8x dynamic resolutions and 2x dynamic batches (Text2Image and Image2ImageControlNet) as well as Segmind-Vega 1024x1024 (Text2Image). Stable Diffusion with Ryzen AI Software 1.7 is also seeing up to a 40% performance improvement for all supported models using the native BFP16 format.
Downloads and more details on Ryzen AI Software 1.7 via GitHub . Windows and Linux installation instructions for this user-space Ryzen AI NPU support can be found via the Ryzen AI documentation .