Performance Check
Not to be confused with the Performance Options options that set the number of sampling steps used during image generation, the Check Performance button is located under the Extras tab, just above the System Information group.
Check Performance has several functions related to NVIDIA video interface support:
- To determine if the video system (GPU) configuration is optimal
- To verify if the video driver fully supports FooocusPlus
- To enable optional GPU reconfiguration after FooocusPlus restart
- To report the speed statistics for the last generative batch
For the purpose of testing, troubleshooting or just curiosity, the Metrics section shown above reports on the speed of the last generative batch. The Generation Time displays the duration in minutes and seconds for the last batch, and Image Quantity shows how many images were in the batch. Lastly, "Throughput" displays the generative speed in two standard metrics, images per minute and seconds per image.
In the Performance Check image above, the Status line provides a quick summary of the graphics system. In this case the user has a non-Blackwell NVIDIA GPU but has used the run_FooocusPlus_cu130.bat file to override the automatic setup. The status is therefore reported as "Experimental Configuration Active".
If the video driver were outdated, the Status line would read "Video Driver Upgrade Recommended", with the details that "Your NVIDIA graphics driver is older than 580.65, the required version. Please upgrade your NVIDIA display driver to version 580.xx or newer to ensure maximum performance and stability."
The Details section elaborates on this configuration. The batch file has made CUDA 13.3 active instead of CUDA 12.8. CUDA, the Compute Unified Device Architecture, is a computer language that runs in the graphics system. CUDA 13 was designed to provide optimal generative performance with Blackwell 50xx GPUs. Testing so far has shown that CUDA 12.8 provides better overall performance when pre-Blackwell NVIDIA GPUs are used. Flux models may run from about 6% to 17% slower with CUDA 13 and SDXL models may operate from 3% faster to 9% slower.
You may have quite different results when testing your own video hardware, and if so please contact the FooocusPlus team to let us know. When testing, I recommend that you generate one image first to ensure that system loading times do not affect the GPU performance results, then run a test batch of about ten images.
If you click on Reconfigure PyTorch the torch identification file will be removed.
In this example, the user chose to Reconfigure Pytorch to restore the recommended CUDA 12.8 configuration. The message above will appear and CUDA 12.8 will be restored at the next startup.
As shown above, at FooocusPlus startup the Python console display will declare that "Torch needs to be installed" then PyTorch and six related libraries will be downloaded and installed. All being well the Performance Check box will list the Status as "Optimal", as shown in the final image.
