Archive — US — turtlesai
Stable Diffusion Runs Faster on AMD GPUs: Optimized Models Arrive
Stability AI and AMD have optimized Stable Diffusion for AMD GPUs and Ryzen AI APUs, achieving up to 3.8x faster image generation. The models, converted to ONNX format, are available on Hugging Face and are compatible with the Amuse 3.0 application, enhancing efficiency in image generation tasks.
Key points
- Collaboration between Stability AI and AMD to optimize Stable Diffusion for AMD hardware.
- Release of ONNX optimized versions, increasing performance: up to 2.6x for SD3.5 models and 3.8x for SDXL models compared to original PyTorch.
- Models available on Hugging Face with '_amdgpu' suffix, including various versions.
- Amuse 3.0 application provides an intuitive interface for users.
- Optimizations include flash attention, weight pruning, and custom fusions, effectively leveraging AMD Radeon RX 9000 and Ryzen AI APU capabilities.
Why it matters
- The move aligns with the 2025 IT Trend of optimizing AI tools for specific hardware, enhancing performance.
- Similar trends observed in GPU advancements by NVIDIA and Intel for machine learning and AI applications.
- Reflects the growing industry focus on partnerships between software and hardware companies to maximize efficiency.
The optimization of Stable Diffusion for AMD hardware marks a significant advancement in generative AI, enhancing performance for developers and users, and setting the stage for future AI model improvements in hardware-specific environments.