Headless jetson nano with hardware accelerated gstreamer and opengl

[blakat360]

How do I run a jetson nano headless but keep hardware acceleration? Specifically, I need hardware accelerated gstreamer and OpenGL.

I’ve seen solutions which use the xf86-video-dummy driver to set up a virtual xserver. This seems to let CUDA do its thing while forcing software rendering in OpenGL. How do I get around this?

I want to render to textures in opengl and then stream said textures over rtp using gstreamer.

[DaneLLL]

Hi,
There is APIs to get EGLImage from NvBuffer and here is a sample:
Nano not using GPU with gstreamer/python. Slow FPS, dropped frames - #8 by DaneLLL

You can check if you can use EGL and OpenGL to implement your usecase. There is demonstration of rendering frames in

/usr/src/jetson_multimedia_api/samples/common/classes/NvEglRenderer.cpp

However, we don’t try this case in headless mode. In headless mode, you should still need to call:

        egl_display = eglGetDisplay(EGL_DEFAULT_DISPLAY);
        eglInitialize(egl_display, NULL, NULL);

        __SOME_PROCESS__;

        eglTerminate(egl_display);