Read the full article here.
Key Takeaways
Stable Diffusion has been upgraded in partnership with AMD, resulting in ONNX-optimized versions tailored for AMD Radeon™ GPUs and Ryzen™ AI APUs. These improvements support models such as Stable Diffusion 3.5 and XL variants, which now operate significantly faster compared to their PyTorch counterparts. The collaboration ensures smoother integrations into various production workflows, ideal for businesses looking to scale their creative processes leveraging AMD’s hardware. The optimized models, which boast up to 3.8 times improvement in inference performance, are accessible on Hugging Face with the "_amdgpu" suffix. For users eager to try out these advancements, the models can be tested via Amuse 3.0. Additionally, AMD's blog provides further insights into the technical specifics of these enhancements.
Key Takeaways
- Stability.ai and AMD have collaborated to optimize select ONNX-compatible versions of the Stable Diffusion models.
- The optimized models are designed for AMD Radeon™ GPUs and Ryzen™ AI APUs, ensuring faster and more efficient performance.
- Versions available include Stable Diffusion 3.5 Large, 3.5 Large Turbo, XL 1.0, and XL Turbo, available on Hugging Face and labeled with "_amdgpu".
- AMD optimized models offer up to 2.6x faster performance for SD3.5 models and up to 3.8x for SDXL models compared to base PyTorch versions.
- These optimizations enable easier integration into existing workflows for faster creative applications.
- You can explore these models using Amuse 3.0 and find more details in AMD’s blog post.
Stable Diffusion has been upgraded in partnership with AMD, resulting in ONNX-optimized versions tailored for AMD Radeon™ GPUs and Ryzen™ AI APUs. These improvements support models such as Stable Diffusion 3.5 and XL variants, which now operate significantly faster compared to their PyTorch counterparts. The collaboration ensures smoother integrations into various production workflows, ideal for businesses looking to scale their creative processes leveraging AMD’s hardware. The optimized models, which boast up to 3.8 times improvement in inference performance, are accessible on Hugging Face with the "_amdgpu" suffix. For users eager to try out these advancements, the models can be tested via Amuse 3.0. Additionally, AMD's blog provides further insights into the technical specifics of these enhancements.
