2023-12-04 01:00:00
The AI QuickSet software suite launched by ASRock includes Stable Diffusion WebUI and supports accelerated graphics performance through Microsoft Olive. According to actual measurements, it can bring 10 times the performance. Installation and model optimization AMD previously posted tutorials on the use of the Stable Diffusion WebUI DirectML branch version on its official blog. You can use the Microsoft Olive tool to convert the original PyTorch format model to the ONNX format and use the DirectML API during calculations. , achieving nearly 10 times the performance. However, the author encountered some technical problems while following the instructions, which cannot be eliminated at present, resulting in the inability to apply performance optimization. Fortunately, the AI QuickSet software suite launched by ASRock greatly simplifies the overall installation process. Users only need to Install the program with the help of , and you can enjoy the advantages of greatly improved performance AI without additional settings. However, it should be noted that the computer requires an ASRock AMD Radeon RX 7000 series graphics card to install AI QuickSet. ASRock officials also stated that the current AI QuickSet program will continue to develop and add more practical AI applications in the future to bring more diverse and convenient AI functions to users. AI QuickSet download location: ▲ According to official data provided by AMD, using the Stable Diffusion WebUI DirectML branch version with a model converted to ONNX format can bring nearly 10 times the rendering performance. ▲ The AI QuickSet software suite launched by ASRock can simplify the installation procedure of the Stable Diffusion WebUI DirectML branch version. ▲ The hardware used in this test is the ASRock Radeon RX 7800XT Steel Legend graphics card. ▲ Readers can download AI QuickSet from the ASRock official website. ▲ The installation procedure is the same as that of ordinary Windows applications. ▲ Taking version 1.1.13 as an example, after the installation is completed, 3 shortcut icons will appear on the desktop. The rendering performance is improved by 10 times. If you want to use the AI rendering environment with optimized performance, you can execute the “Launch Stable Diffusion WebUI ONNX” shortcut. However, it should be noted that the current DirectML version only supports the use of Stable Diffusion 1.5 basic models. The Olive model conversion tool in the web interface cannot be used to convert other Checkpoint models (the program specifies model files in Hard coding mode), and it does not yet support LoRA. , there are many functional limitations. If you want to use other models, you can execute the “Launch Stable Diffusion WebUI” shortcut. After opening the DirectML version of the Stable Diffusion WebUI interface, you need to select “stable-diffusion-v1-5-olive” from the Stable Diffusion checkpoint drop-down menu in the upper left corner. [Optimized]”, and the rest of the operations are the same as the normal version. The author used ASRock Radeon RX 7800XT Steel Legend for testing. In the process, the Batch size was set to 1 and 4 respectively, and the Batch count was fixed at 1. After performing 2 rounds of testing to ensure that the test results had no extreme values, the average was taken as the test score. It can be seen from the results that after optimization, the performance of the AMD camp graphics card is close to that of the 768 x 768 resolution. When the Batch size is 1, one image is calculated at a time. The performance of the DirectML version is about 4.64 times that of the general version. . If the Batch size is set to 4 and 4 images are calculated at the same time, the speed of the DirectML version will not be greatly affected, but the speed of the general version will be seriously reduced, resulting in a performance gap of 45.29 times. ▲ When executed for the first time, the program will automatically install the required files and convert the model, which will take a long time. ▲ After the second execution, the opening speed will return to normal. ▲ When executing, you need to select “stable-diffusion-v1-5-olive” from the Stable Diffusion checkpoint drop-down menu in the upper left corner [Optimized]” to apply the optimization model. ▲ Parameter settings and execution results of the DirectML version. ▲ When the Batch size is set to 1, the performance of the DirectML version is about 4.64 times that of the general version. The Olive tool from the AMD camp can bring effective performance improvements, but compared with the TensorRT optimization tool provided by NVIDIA, it is slightly less convenient and practical, and it also needs to be updated and improved in the future. (Return to the series of articles)
1701667069
#Stable #Diffusion #Graphics #User #Manual #Handson #play #ASRock #QuickSet #optimizing #AMD #graphics #cards #increase #graphics #performance #times