Support of CUDA 12 in HALCON 22.11 Steady

This patch enables you to use CUDA 12 in your HALCON 22.11 Steady installation and thus use the latest generation of Nvidia GPU cards.

Now there are 2 packages available - one for Windows and one for Linux:

  • Linux: tensorrt_fix_2211s_linux.tgz
  • Windows:

Basically the workaround steps for the tensorrt issue in HALCON 22.11 Steady are for both operating systems similar (but not the content of the packages):

  1. Rename thirdparty folder located in your HALCON installation (%HALCONROOT%\bin\x64-win64\thirdparty\cuda11_1) from cuda11_1 to something else like cuda11_1_ori.
  2. Extract the corresponding package directly into the %HALCONROOT% directory.

Known issues:

  • There could be regressions of the gpu usage of some models, and variations of the inference performances.
  • Runtime initialization i.e. optimize_dl_model_for_inference takes longer (cudnn 8.9) vs original 22.11 (cudnn 8.2).

For downloading the files you need the following credentials:

User: ftp

Password: vdQ.I}7_

Here you find the related files: