The Procyon AI benchmark measures inference times on popular models like MobileNet, ResNet, and YOLO. It's designed to show how well a system can accelerate machine learning tasks in real use.
The M1 Pro scores 130 overall, close to the A9 Max's 137. Model inference times include 1.08 ms (MobileNet), 10.48 ms (ResNet-50), 30.3 ms (Inception V4), 37.0 ms (DeepLab V3), 74.9 ms (YOLO V3), and 3,412 ms (Real-ESRGAN). These results show that Arc's AI and compute support is viable for many inference tasks.