Usefulness of RadeonHD GPU in a headless server

[Havin_it]

Hi,

So I've been wondering about this for a while and it's finally reached the top of the pile. Back in January I got an HP Microserver: great little workhorse, but it does seem odd for it to have a RadeonHD graphics card (albeit a "Mobility" one so not really a monster I guess) when it's very much a headless box by design.

Anyhow, I'm just wondering whether I can leverage it in any fun ways through the wonders of Gentoo (or maybe I already am without knowing it?). For example, I use TigerVNC (client and server) to work on the box. I've heard that this does make some use of GPU acceleration, but not sure whether that applies when not using it as an Xorg module (I haven't really tried that as I'm not sure how (or whether) that's supposed to work in a headless setup). Tiger suffers from a paucity of docs/faq, so just wondering if anyone knows more details about this aspect.

Also there is the business of CUDA/OpenMP/whatever, i.e. tapping into the GPU for more general-purpose processing. This seems a mess of competing standards and "in the works" development at the moment: is there anything much that can be achieved with it at the moment?

I've never owned a machine with a "name-brand" (i.e. not Intel) GPU before, so I feel I ought to be getting the maximum benefit from it somehow. TIA for any enlightenment on the subject.

[Ant P.]

Nothing useful yet, but in a few months' time, OpenCL should start being usable in Mesa.

[chithanh]

The RS780 IGP in the Microserver is not OpenCL capable.

With recent GPU offloading work, you may be able to use some 3D accelerated VNC desktop in the future. Also if/when UVD support comes, it may be possible to do accelerated video decoding on the server.

[Havin_it]

Thanks for the replies, both of you. OpenCL, that's the one I was thinking of - too many damn acronyms/buzzwords around this subject!

UVD is a new one on me. I read up a little (Wikipedia) and it sounds interesting but could be a long wait to see it bear fruit.

Thinking about it, I guess the normal X acceleration benefits would not apply to VNC, since the GPU (or IGP, if I must ) does what it does as a middle-point between the CPU and a physical screen (which it outputs directly to, since the video-out ports are attched to it). I guess then that it's not intended to redirect that output back into the CPU/main memory so it can hand it off to the VNC server, and if it even can do that it's less efficient. Am I understanding accurately?

I read a bit about VirtualGL, but I'm not clear at all on what role that plays in TigerVNC (if any at this moment - I know it's in TurboVNC but I have other reasons for wanting to use Tiger). I don't think Tiger uses Mesa or any of that, unless perhaps it's used as an Xorg module (which it's not here since there's no monitor).

So I'm assuming that there's no reason for me to set any USE flags such as opengl, dri, drm etc on this setup.