hybrid-graphics-linux team mailing list archive

[Pigeon <pigeon@xxxxxxxxxxx>]

> Question 1:  is the cpu encoding the rendered buffer a bottleneck?

	When using VGL_COMPRESS=jpeg it does use
a lot more cpu, hence a bottleneck. My feeling is jpeg compress is only
useful when the data is actually going thru network.

	You could also use VGL_PROFILE to find out a bit more.


> Question 2:  is it possible to do encoding on the GPU?

	Possibly. I haven't looked at VirtualGL's code. But you can if
you use YUV transport, one could in theory use a shader to
encode/decoder YUV.


> Question 3:  Is the intel card decoding using hardware?

	I tried using VGL_COMPRESS=xv, and I assume on the intel card
it would be using XV (hence hardware) to decode and display the image.
It seems to give pretty good performance too. Though I personally mind a
little myself when using yuv or xv since the color is washed out a
little due to the nature of the format.

	I also did notice when I use vglclient -gl, it does give a
lower cpu usage comparing without -gl.



Pigeon.