very slow performance of glxgears (68 fps)

Jim Gettys jg at freedesktop.org
Wed Feb 18 07:23:09 PST 2009


On Wed, 2009-02-18 at 15:06 +0000, John Tapsell wrote:
> 2009/2/18 Corbin Simpson <mostawesomedude at gmail.com>:
> > John Tapsell wrote:
> >> 2009/2/18 Matthias Hopf <mhopf at suse.de>:
> >>> On Feb 06, 09 01:56:19 +0000, John Tapsell wrote:
> >>>>>>> glxgears is not a benchmark.
> >>>>> At openSUSE we print out a warning now (well, this change went into
> >>>>> *after* 11.1, unfortunately), that this is not a benchmark. We got really
> >>>>> tired of these statements.
> >>>> Except that it _is_ a benchmark.  Or rather waas before the change.
> >>> No, it never was.
> >>>
> >>>> If for example, without vsync, a user gets just 100fps on their new
> >>>> nvidia card, then that is clearly showing that something is wrong.
> >>> So it is a validation tool, but not a benchmark.
> >>>
> >>> A benchmark - by definition - gives you performance figures that can be
> >>> compared with other systems, which gives you a reasonable notion of
> >>> which of the systems is better.
> >>
> >> Right.  If one system gets 100fps and another gets 1,000fps, then you
> >> compare the two systems and say that one is better than the other.
> >> All you are saying is that it's not an accurate comparison if the
> >> numbers are close.
> >
> > A benchmark also measures specific hardware performance and tests
> > optional features. glxgears only tests clear speed and the rate at which
> > a modest-sized vertex buffer can be rendered. A useful sanity test, but
> > nothing more.
> 
> Well whatever we call, it's no longer useful for sanity checking if
> it's limited to 60fps  :-)

Heh.  This means you can tell we're finally dealing with vertical
retrace in a sane way, so others may disagree ;-).
                        - Jim

-- 
Jim Gettys <jg at freedesktop.org>




More information about the xorg mailing list