So, AMD APP support for GPUs on Linux without a running X server has been brought up in this thread multiple times. Could we get an update from AMD on whether or not this is being worked on, or even considered by the driver team?
We do a lot of research on heterogeneous computing, especially GPU acceleration of scientific code. So far, we have primarily used NVidia/Intel hardware, but would like to also explore AMD hardware, especially the Fusion architectures and high-end Radeon cards. However, our experience so far with AMD hardware has been less than ideal since we need to majorly modify the software installations on our compute machines to accomodate a full X server installation along with appropriate permissions and handling of the DISPLAY environment variable depending on the type of application we are running.
Simply put, requiring a running X server on headless compute nodes not only needlessly complicates the software installations, but requires us to carefully balance GPU usage against system RAM usage from the X server.
