Procyon AI Image Generation

Procyon AI Image Generation

Benchmark GPU AI Image Generation Performance

The Procyon AI Image Generation Benchmark provides a consistent, accurate, and understandable workload for measuring the inference performance of on-device AI accelerators. This benchmark was developed in partnership with multiple key industry members to ensure it produces fair and comparable results across all supported hardware.

The benchmark includes three tests for measuring the performance from low power NPUs to high-end discrete graphics cards. 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 moderately powerful discrete GPUs, we recommend the Stable Diffusion 1.5 (FP16) test. Finally, we designed the Stable Diffusion 1.5 (INT8) test for low power devices using NPUs for AI workloads.

The Procyon AI Image Generation Benchmark can be configured to use a selection of different inference engines, and by default uses the recommended optimal inference engine for the system’s hardware.

Buy now

Features

  • A range of tests built around an image generation workload, using state-of-the-art neural networks.
  • Designed to measure the inference performance of a wide range of AI accelerators.
  • Benchmark with NVIDIA® TensorRT™, Intel® OpenVINO™, and ONNX with DirectML.
  • Verify inference engine implementation and compatibility.
  • Simple to set up and use via the Procyon application or via command-line.
  • Test with multiple versions of the Stable Diffusion AI model.
  • Compare up to 4 results side-by-side in the app.

Benchmark details


Stable diffusion, released in 2022, made using AI for text-to-image generation on their own hardware accessible for the everyday consumer. Given its ease of access, wide usage, and creative aspect, text-to-image generation quickly became one of the most memorable AI use cases for the public.

The AI Image Generation Benchmark uses a set of standardized text prompts for a reliable and consistent AI image generation workload. Results provide an overall score for easy comparison, as well as further detailed scores and the generated images for closer inspections of performance and quality.

Test Workload Image resolution Batch size Steps
Stable Diffusion XL (FP16) Heavy 1024 x 1024 1 100
Stable Diffusion 1.5 (FP16) Medium 512 x 512 4 100
Stable Diffusion 1.5 (INT8) Light 512 x 512 1 50

Results and insights


Benchmark scores

Compare AI Inference performance with two different versions of the Stable Diffusion model. procyon AI Image Generation benchmark scores

Detailed scores

Inspect generated images, and get detailed scores for each image generation batch. procyon AI Image Generation detailed scores

Hardware monitoring

Get detailed metrics on how CPU and GPU temperatures, clock speeds and usage change during the benchmark run. procyon AI Image Generation hardware monitoring

Developed with Industry expertise


Procyon benchmarks are designed for industry, enterprise, and press use, with tests and features created specifically for professional users. The Procyon AI Image Generation Benchmark was designed and developed with industry partners through the UL Benchmark Development Program (BDP). The BDP is an initiative from UL Solutions that aims to create relevant and impartial benchmarks by working in close cooperation with program members.

Inference Engine Performance

With the Procyon AI Image Generation Benchmark, you can measure the performance of dedicated AI processing hardware and verify inference engine implementation quality with tests based on a heavy AI image generation workload.

Designed for Professionals

We created our Procyon AI Inference Benchmarks for engineering teams who need independent, standardized tools for assessing the general AI performance of inference engine implementations and dedicated hardware.

Fast and easy to use

The benchmark is easy to install and run—no complicated configuration is required. Run the benchmark using the Procyon application or via command-line. View benchmark scores and charts or export detailed result files for further analysis.

Procyon AI Inference Benchmark

Free trial

Request trial

Site license

Get quote Press license
  • Annual site license for Procyon AI Image Generation Benchmark.
  • Unlimited number of users.
  • Unlimited number of devices.
  • Priority support via email and telephone.

Benchmark Development Program

Contact us Find out more

Contact us


The Benchmark Development Program™ is an initiative from UL Solutions for building partnerships with technology companies.

OEMs, ODMs, component manufacturers and their suppliers are invited to join us in developing new AI processing benchmarks. Please contact us for details.

Minimum system requirements

OS Windows 10, 64-bit or Windows 11
Processor 2 GHz dual-core CPU
Memory 16 GB
Storage 20 GB (75 GB recommended)

Stable Diffusion XL requirements

TensorRT NVIDIA RTX GPU with 10GB VRAM
OpenVINO Intel Arc discrete GPU with 16GB VRAM
ONNX Runtime 16GB VRAM

Stable Diffusion 1.5(FP16) requirements

TensorRT NVIDIA RTX GPU with 10GB VRAM
OpenVINO Intel Arc discrete GPU with 8GB VRAM or Intel integrated GPU with 32GB system RAM
ONNX Runtime 8GB VRAM (Discrete GPU) - 32GB RAM (Integrated GPU)

Stable Diffusion 1.5 (INT8) requirements

TensorRT NVIDIA 30 series GPU or later
OpenVINO An Intel NPU, Intel integrated GPU or Intel Arc discrete GPU

Support

Latest 1.0.142 | September 16, 2024

Languages

  • English
  • German
  • Japanese
  • Portuguese (Brazil)
  • Simplified Chinese
  • Spanish