Does bamboo disable the GPU/rendering when running commands?

[AmitB]

I am running unit tests in bamboo as a C++ exe in a script configuration. I added GPU/rendering capabilities to my tests, and they are always crashing in bamboo. If I log into the server directly (not as remote desktop), the tests run fine. Is this a limitation of Bamboo? Any workarounds?

[AmitB · Accepted solution]

Looks like the solution is to use a Quadro 4000 or tesla gpu, as they can run rendering tasks in a remote desktop enviroment.

[LucasA]

Are you trying to make use of a generic OpenCL supported card or some specific sollution (nVidia Fermi for example)? Note that if you are using a generic card and making no use of CUDA framework (assuming that you are using a nVdia solution), you'll need to load the GPU driver and all the necessary drivers, since remote agent will not open a graphical session, the GPU will not be used. Refer to the solution provider manual and try to reproduce your plan without Bamboo. If you check all the steps and still not work with Bamboo, please raise a support ticket at http://support.atlassian.com.

Lucas Lima

Atlassian Support

[AmitB]

I am using CUDA, and the code is running fine when I run locally on the server. I am also using OpenGL for offline rendering, then passing on the data to CUDA for processing. Could the problem be related to OpenGL?

I only get the error when running in Bamboo. Do I need to do something extra for CUDA in order to enable the GPU via CUDA?

Is there a way to force loading of necessary display drivers?

[Przemek Bruski]

If your tests require acces to GUI in any way, you need to configure autologin for you agent. See here:

https://confluence.atlassian.com/display/BAMBOO/Creating+a+custom+elastic+image+-+Windows , I think the folded section about "not headless" builds applies to your case.

[AmitB]

They don't require GUI, but access to GPU for offline rendering (OpenGL). I followed your instructions, but I still get a crash when attempting to access the GPU.