Procyon: the creators of 3DMark release an AI benchmark

0

With a toolbox including 3DMark, PCMark and VRMark, UL Solutions couldn’t ignore the advent of AI. With Procyon, they have. This benchmark will enable you to measure AI image generation performance on your PC. According to UL, this benchmark has been designed with the help of industry members to ensure fair and comparable results across all supported hardware.

Procyon

With the proliferation of announcements around AI ( Intel, Microsoft…), measuring AI inference now requires a selection of benchmarks to optimally measure the AI inference performance of all available consumer hardware. Needless to say, like VR in its day, we’re going to have to really test our PC’s AI management capabilities.

The Procyon benchmark explained in detail:

UL Procyon AI Image Generation benchmark
To better measure the performance of mid-range and high-end dedicated graphics cards, this benchmark contains two tests performed using different versions of the Stable Diffusion model, and we hope to add more tests in the future to support other performance categories.

The Stable Diffusion XL (FP16) test is our most demanding AI inference workload, and only the latest high-end GPUs meet the minimum requirements to run it. For mid-range discrete GPUs, Stable Diffusion 1.5 (FP16) is the recommended test.

Procyon benchmark IA
Test the performance of several AI inference engines
Like our AI Computer Vision Benchmark test, you can freely switch between several leading inference engines, allowing you to compare differences in engine performance, hardware using the same engine or the best implementations of AI inference performance on different devices. By default, the benchmark selects the optimal inference engine for the system hardware.

Currently, the AI image generation benchmark supports the following inference engines. We plan to add support for other inference engines in the future to provide optimal performance for all supported AI hardware:

ONNX® runtime with DirectML™
NVIDIA® TensorRT™
Intel® OpenVINO™

The benchmark is available for a fee. However, we’re going to take a quick look at it to offer you some interesting features.