UL Procyon AI Image Generation Benchmark for Windows version history

A benchmark version number is specific to a test. Benchmark version numbers change rarely and only when absolutely necessary to accommodate changes in third-party applications or bug fixes.

UL Procyon AI Image Generation Benchmark for Windows v1.1.157

December 09, 2024

Please note that updates to inference engines runtimes can impact results. We recommend you benchmark your devices again after an inference engine runtime update for up-to-date performance results.


Fixed

  • Fixed an issue in the AI Image Generation Benchmark where the benchmark would sometimes fail to end correctly when using the OpenVINO runtime. 

UL Procyon AI Image Generation Benchmark for Windows v1.1.152

November 26, 2024

Please note that updates to inference engines runtimes can impact results. We recommend you benchmark your devices again after an inference engine runtime update for up-to-date performance results.


Updated

  • Updated the Intel OpenVINO inference engine to version 2024.5
  • Updated the ONNX runtime with DirectML inference engine to version 1.20.0. DirectML EP now uses the compute command buffer for inference preventing TDR issues on lower-end hardware.

UL Procyon AI Image Generation Benchmark for Windows v1.1.144 

September 16, 2024

This update improves the TensorRT engine build time on first benchmark runs. These updates may impact results for systems using NVIDIA hardware. We recommend you benchmark your devices again for up-to-date performance results.


New

  • Updated the AI Image Generation workload version to 1.1.144. This adds support for NVIDIA timing caches, improving the speed of building the TensorRT engine on the first run.     

Fixed

  • Fixed an issue in the AI Image Generation Benchmark where some systems would fail to build the TensorRT engine or perform poorly on the first run using Stable Diffusion 1.5.
  • Fixed an issue in the AI Image Generation benchmark where the benchmark would fail during the model initialization step when using Stable Diffusion 1.5 and the TensorRT inference engine.
  • Fixed an issue in the AI Image Generation Benchmark where the benchmark would occasionally fail when run offline.

UL Procyon AI Image Generation Benchmark for Windows v1.0.139

September 6, 2024

This update supports the release of the integer quantized (INT8) UL Procyon AI Image Generation Benchmark. Please note that there are updates to inference engines runtimes that may impact results. We recommend you benchmark your devices again after an inference engine runtime update for up-to-date performance results.


New

  • Added support for INT8 quantization for Stable Diffusion 1.5 to the Procyon AI Image Generation Benchmark. This currently supports Intel NPUs, iGPUs and Intel Arc discrete GPUS using OpenVINO, and NVIDIA RTX 30 series GPUs and later using TensorRT.

Updated

  • Updated the ONNX runtime with DirectML inference engine to 1.19 for the Procyon AI Image Generation Benchmark  
  • Updated the TensorRT inference engine to version 10.3 for the Procyon AI Image Generation Benchmark

Fixed

  • Fixed an issue where the Stable Diffusion 1.5 model used in the Procyon AI Image Generation Benchmark was not available. It now uses an alternative Stable Diffusion 1.5 model (nmkd/stable-diffusion-1.5-fp16). This does not affect results Please note that manually installed Stable Diffusion Models need to be removed first, as this new model has a different folder structure.  

UL Procyon AI Image Generation Benchmark for Windows v1.0.126

August 26, 2024

This version updates the Intel OpenVINO runtime. We recommend you benchmark your devices again after an inference engine runtime update for up-to-date performance results


Updated

  • Updated the Intel OpenVINO inference engine to version 2024.3 for AI Image Generation Benchmark.

Fixed

  • Fixed the VAE step in the AI Image Generation Benchmark to run at FP16 precision when using OpenVINO inference engine, for both Stable Diffusion 1.5 and Stable Diffusion XL. 

UL Procyon AI Image Generation Benchmark for Windows v1.0.76

March 25, 2024

  • Launch version