2 Replies Latest reply on Dec 28, 2016 6:00 PM by Intel Corporation

    Remote Linux Workstation

    JayHilliard

      I'm interested in the best offerings from Intel for remote visualization of OpenGL apps on RHEL.

      The proprietary nature of GRID isn't compatible with the need for us to deliver our scientific linux opengl workstations remotely. Many applications require GLX/OpenGL.  Solutions tested:

       

      Teradici - Hardware solution - works very good, top tier.

      HP RGS - tier 2, unstable, bugs when OS revs.

      Nvidia Grid - Expensive, proprietary, and we can't install vendor graphics libraries in the target host.

      Cendio Thinlinc (Xvnc) with VirtualGL - Works, and folks can share a gpu this way, but performance needs much improvement. A bit hacky.

      Xrdp/x11rdp - Not better than any of above, in my experience, yet. I understand VirtualGL can be used with it?

       

      This is further complicated by the eventual switch from X to Wayland/weston, and the loss of some of the options above.

       

      In a perfect world, the discrete graphics vendor would provide a hardware accelerated remote method as an alternative to DisplayPorts, or even in place of Display Ports.

       

      I Imaging a Xeon Phi system where I can assign a number of bootable VPU's to be a host, some to process OpenGL/compositing for that host, and the ability to have more that one of these "not so virtual" hosts existing on a single Xeon Phi based system and the ability to keep it in a rack and roll out thin clients. At the same time, some of the the Phi cores go be a great test bed for us to explore VPU processing in our own software.

       

      My scenario is currently Nvidia Graphics, RHEL7, lightdm/Mate 1.16 as the Desktop, and the intel ParallelStudio 2017 stack.  I know there may be a "step down" in graphics performance, but the trade-off might be worth it if it can be close to the top tier (Teradici) and better than the 2nd Tier.

       

      I'm just throwing this out there, wondering if this type of thing is on Intel's radar, and if so, I would like to be involved.