AMD's advancements in the AI field are noteworthy, particularly its recent optimizations for the Stable Diffusion model. Stability AI recently released an ONNX-optimized version of Stable Diffusion, resulting in significantly improved performance for AMD Radeon graphics cards and Ryzen integrated graphics in AI tasks – up to 3.8 times faster. This progress not only narrows the gap with NVIDIA in ecosystem optimization but also provides tangible benefits to AMD users.
Over the past year, AMD has collaborated closely with numerous partners, including OS vendors (OSV), original equipment manufacturers (OEMs), and independent software vendors (ISVs), focusing on hardware design, drivers, compilers, and machine learning models to enhance the AI application performance of AMD hardware. According to AMD's official data, after a series of optimizations, the inference performance of SDXL1.0 and SDXL Turbo on AMD hardware has improved by 3.8 times compared to the basic PyTorch model, while SD3.5's performance has increased by 2.6 times.
In practical applications, the RX 9070XT graphics card running SDXL1.0, SDXL Turbo, SD3.5Large, and SD3.5Large Turbo models saw efficiency improvements of 3.1 times, 1.5 times, 3.3 times, and 2.1 times, respectively. Simultaneously, the Ryzen AI MAX+395 graphics card showed performance improvements of 1.4 times, 3.3 times, and 3.0 times for SDXL1.0, SD3.5Large, and SD3.5Large Turbo, respectively. These performance gains make AMD graphics cards excel in image and video generation applications.
Furthermore, AMD has specifically highlighted the optimized Stable Diffusion models, using the "_amdgpu" suffix (e.g., stable-diffusion-3.5-large_amdgpu). Users can download these optimized versions from Stability AI or Hugging Face. The jointly developed Amuse3.0 version by AMD and TensorStack AI is also available, offering a smoother user experience. It's important to note that to successfully use these optimized models, AMD GPU drivers version 24.30.31.05 preview or higher, or the upcoming 25.4.1 official release, is required.
These advancements not only enhance the AI capabilities of AMD graphics cards but also provide developers and users of AMD hardware with more choices and flexibility.
