Clij headless benchmarking errors

My colleagues just found out that the “Remote Desktop Session Host” program on the server 2019 (and apparently 2016) is causing the issue.
Deinstalling it allows clij to work again, but deactivate the remote desktoping functionality.

@bramvdbroek (and others) I will get back to you if we find some alternative :wink:
Thanks @haesleinhuepf for the time spent investigating this issue !

Hello everybody.
I can confirm that indeed the RD Session Host causes the issue with CLIj.
Deactivating it, restores the functionality of CLIj but restricts the remote desktop session to 2 on a windows 2019 server.

1 Like

Hey Ioannis, @ialexop,

we do have a potential solution for this problem here:

Would you mind checking if it works on your system?


Dear Robert,
Thanks for pointing out this.
I already checked this version of cuda toolkit (without the Visual Studio though) before I uninstall the RD Session Host role, but without any luck.
The CLIj worked like a charm when I was working using a remote desktop connection (but without the RD Session Host). When I installed back the Host, I got back the error that is described here.
I will try more on Monday installing also the Visual Studio.
Kind regards

PS: I have a very similar configuration as the HIVE system with Windows 2019 Server and 3x RTX6000

1 Like

Wow, that sounds like an amazing system! CLIJ in ImageJ macro doesn’t support multiple GPUs. But if you want to dive into this, I’m happy to introduce you to the CLIJ Falcon Heavy project :wink:

Let me know if the workaround above works and have a great weekend!


Hi Robert,
I would be thrilled to learn more! :slight_smile:
I hope we will manage to fix the issue with the RD Session Host.
Thanks and best

Ah, we’re working in two separate threads. I’m only seeing this last part now.

From here on the subject changes to (mostly) Windows 2019 issues (and the ~1 Tb memory limit error).
Would it be possible to somehow merge it with this thread? Also nice for future reference.
Who has the power to do so? (@haesleinhuepf?)