[18:08 Wed,24.May 2023 by Thomas Richter] |
Microsoft has unveiled a new free tool called Olive that allows many AI models, such as Stable Diffusion, to automatically use hardware acceleration from graphics cards such as NVIDIA&s RTX Tensor Cores to significantly speed them up. Olive (ONNX Live) is a sophisticated model optimisation tool that integrates seamlessly with DirectML to enable hardware acceleration in Windows. ![]()
It compresses, optimises and compiles existing AI models and adapts them to the existing hardware. However, adapted drivers for the respective GPU are necessary for maximum performance so that it can be used optimally. Fortunately, NVIDIA has today published corresponding new ![]() How much do the optimisations bring?According to the ![]() ![]() A special version of Stable Diffusion, optimised by Olive, is necessary for this leap in performance, but it can be compiled with a little expertise github.com/microsoft/Olive/tree/main/examples/directml/stable_diffusion. However, thanks to the free availability of Olive, which was published under the MIT licence, user-optimised versions of Stable Diffusion will probably appear very soon for less experienced users. At the moment, support for fine-tuning Stable Diffusion using LoRA is only experimental. Who can use Olive?The new Game Ready drivers are available for both the current GPU generation and very old NVIDIA graphics card series - starting with the GeForce GTX 700 - according to devblogs.microsoft.com/directx/ dml-stable-diffusion/ (Microsoft) but only the newer RTX 30 and RTX 40 series benefit from Olive DirectML optimizations through the use of Tensor cores. Apart from that, of course, the limitations of Stable Diffusion regarding the size of the graphics card memory required for the different models also apply. ![]() However, owners of AMD&s graphics cards won&t go away empty-handed - they just have to be patient. So ![]() How does Olive work?Stable Diffusion is a complex machine learning model that includes several smaller models. These models were originally developed in the popular PyTorch Machine Learning Framework. Microsoft&s Olive converts these PyTorch models into another format called ONNX and then performs several complex optimization steps that make the models more efficient when later run. These can then also be optimized for special hardware such as NVIDIA&s Tensor Cores, which can then be run on the PC or in the cloud and again bring significant speed increases. ![]() deutsche Version dieser Seite: Microsoft Olive: Neues kostenloses Tool verdoppelt Performance von Stable Diffusion |
![]() |