--=-Y7+qe8teLemYBuhRRven
Content-Type: text/plain
Content-Transfer-Encoding: quoted-printable

On Thu, 2007-03-22 at 20:36 +0100, Matthias Drochner wrote:
> yhardy@uj.ac.za said:
> > > The "glxgears" score is actually a bit less than with software
> > > rendering
> > Did you turn off debugging?=20
>=20
> The code is compiled in, but the "hw.dri.debug" sysctl flag
> is switched off.
>=20
> Checked a bit more throughly and found that software is only faster
> with the (relatively small) default window size of glxgears.
> At full screen size, dri wins: I'm getting 80 fps; about 9%
> of a CPU (of 2 on a dual-core) is used plus ~10% system time.
> Software rendering gives 64 fps and uses 99% of a CPU.
> The worst combination is btw to have dri enabled but AIGLX
> disabled. It uses a different way of software rendering
> appearently which causes flickering output and gives 19fps
> at 99% CPU load. In case it matters, the driver modules are
> from Mesa-6.5.2.

This sort of stuff is one of the reasons that every DRI developer ever
will yell at you and say "glxgears is not a benchmark!" :)  It's not a
representative workload of 3d activity that needs hardware acceleration,
and furthermore doesn't use normal techniques that any
performance-sensitive application uses.

Some have actually gone so far as to make glxgears require that you pass
-iacknowledgethatglxgearsisnotabenchmark to get the fps numbers.

--=20
Eric Anholt                             anholt@FreeBSD.org
eric@anholt.net                         eric.anholt@intel.com


--=-Y7+qe8teLemYBuhRRven
Content-Type: application/pgp-signature; name=signature.asc
Content-Description: This is a digitally signed message part

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (FreeBSD)

iD8DBQBGEsSlHUdvYGzw6vcRAu9nAJ9LbVNFcPchn528qycbQPOp3YJ6pACfcqWg
g5bRhIZ3bv4/Eqz4Wx780uI=
=a6io
-----END PGP SIGNATURE-----

--=-Y7+qe8teLemYBuhRRven--