intel driver gets slow as soon as I rotate on a 1920x1200 display

Jesse Barnes jbarnes at
Thu Jul 26 14:15:08 PDT 2007

On Thursday, July 26, 2007 1:43 pm Lukas Hejtmanek wrote:
> Hello,
> On Mon, Jul 23, 2007 at 08:54:54AM -0700, Jesse Barnes wrote:
> > > I just commited a typo fix for this, but I haven't a 915/945
> > > mobile by hand to test with fbc enabled. Could you test with
> > > latest git?
> >
> > Thanks Zhenyu, I did a quick scan last Thu. and didn't see
> > anything, this should help.
> >
> > And currently, on 965, I've marked tiling as unsupported, which
> > means framebuffer compression isn't enabled by default.  You can
> > still force it on using "Tiling" "true" and
> > "FramebufferCompression" "true", but there are known rendering
> > bugs; hopefully I can fix those soon.
> I just found that the new memory allocation code forces either all
> buffers to be tiled or none buffer is tiled. The previous version
> tiled at least back and depth buffers.
> Now, if I use tiling, I got glxgears at 990FPS and ppracer at 33FPS.
> If none buffer is tiled then I got glxgears at 200FPS and ppracer at
> 3.6FPS.
> So, I guess this is the issue with slow down.

Yep, not tiling really hurts performance.  The idea is to fix the 
rendering bugs so we can just enable tiling in all cases for all 
buffers.  That should keep performance up and power consumption down, 
and cut down on the number of different configuration options and 
conditionals we have to deal with.  If the rendering stuff isn't sorted 
out by the time we're ready for the next release, I'll kludge the code 
back into its old state, allocating tiled buffers for everything but 
the front buffer, and disabling framebuffer compression by default.


More information about the xorg mailing list