Re: Fwd: How will cairo influence GTK's performance?

[Billy Biggs <vektor dumbterm net>]

Richard Stellingwerff (remenic gmail com):

> > Is there a way you could create a programmatic benchmark (or a command
> > line that does not require user interaction)?
> 
> User perception. But the difference is so huge, that I can tell with
> absolute certainty that it's there. I just can't give you any numbers
> to show much faster it is.
> 
> Things I notice it with are most notably Mozilla. With ati-drivers and
> nvidia-glx, resizing the Mozilla window is pretty slow, ie. the
> window's contents lag from its border very notably. Scrolling a huge
> page like slashdot (comments section) skips a lot, and just feels
> sluggish overall.
> 
> With the DRI drivers however, resizing the window is a LOT smoother,
> and scrolling trough the entire page is smooth as silk, no matter how
> hard I pull that scrollbar :P

  Resizing mozilla tests your window manager and mozilla's rendering
more than GTK+ itself.  However, since you're seeing differences between
drivers, you might want to look at raw throughput to the card.  Compare
the output of this command (let it run to completion, bonus points: run
without a window manager and let it be the only client):

    x11perf -shmput500

  On a 24-bit display, you should see numbers over 200.  If not, it may
be a bug with your kernel with DMA transfers (which would be used more
by the proprietary drivers than the open source ones).

  -Billy