Stable diffusion gets (much) faster on AMD | | | | Turtles AI

  New guide unlocks up to 9.9x faster #StableDiffusion on #AMD #GPUs using model optimizations from #Microsoft #Olive and #DirectML. A new guide (click here) details how users can enable optimized Stable Diffusion performance with Microsoft's Olive tool on AMD GPUs. Microsoft and AMD collaborated to streamline Stable Diffusion, an AI image generation model, leveraging AMD GPU AI capabilities through Microsoft's DirectML API. The optimizations convert the PyTorch model to ONNX format, fuse subgraphs, and quantize from FP32 to FP16, reducing memory footprint and boosting performance. Microsoft Olive handles the optimization pipeline, simplifying tuning of the complex Stable Diffusion model. The guide walks through installing dependencies like Git and Anaconda, then using Olive to generate an optimized ONNX model from Stable Diffusion code. It explains integrating the model into Automatic1111's WebUI and selecting it to run Stable Diffusion inference on AMD GPUs with DirectML. Benchmarks on an AMD Radeon RX 7900 XTX show the optimizations provide up to 9.9x higher iterations per second over default PyTorch. The guide enables AMD #GPU owners to tap into AI performance gains from Microsoft and AMD's collaboration. Highlights: - Convert and optimize Stable Diffusion with Microsoft Olive for AMD GPUs. - Integrate optimized model into Automatic1111 WebUI with DirectML. - Achieve up to 9.9x faster iterations per second over default. This collaboration between Microsoft and AMD promises exciting AI performance gains for AMD GPU owners. With these optimizations, Stable Diffusion can take full advantage of AMD hardware capabilities. We'd love to hear your experiences applying the guide and any speed improvements observed. Let us know if you have any difficulties with the process so we can provide tips. Discuss in the comments!