UL, the company responsible for the 3DMark benchmark for testing central and graphics processors, introduced the Procyon AI Image Generation Benchmark program – it is designed for testing artificial intelligence accelerators.
The program is not included in the 3DMark package intended for consumers, but is aimed at commercial enterprises that have the funds for these needs – the benchmark costs $5,000. It allows you to test the performance of systems in image generation using one of the most popular Stable Diffusion models. For high-performance accelerators, the latest version of the Stable Diffusion XL model is used, which creates images in a resolution of 1024 × 1024 pixels; and mid-range GPUs are tested using the older SB 1.5 platform with a resolution of 512 × 512 pixels.
To ensure unbiased testing and optimal performance on modern hardware, the benchmark supports the most popular APIs: ONNX with DirectML (primarily for AMD Radeon GPUs), Nvidia TensorRT (for Nvidia RTX) and Intel OpenVINO (for Intel graphics). This test is not aimed at consumers, so even among enthusiasts it probably will not be used. But perhaps it will be used in technical reviews of expensive professional equipment.
If you notice an error, select it with the mouse and press CTRL+ENTER.