Re: [Dri-devel] Quake3 slowness?

[Daryll S. <da...@pr...>]

On Tue, Mar 07, 2000 at 07:49:02AM -0800, David Raufeisen wrote:

> I've noticed that the 3.3.x ones were quite faster as well.
> 
> I know that if i use mesa 3.2 quake3 is a lot slower then 3.1(i think).
> 
> Could the Mesa version be the problem?

In general, benchmarking is hard. If you're going to do a benchmark,
make sure you're making as fair a comparison as possible. Under the DRI
there are a few things to be aware of.

      setenv FX_GLIDE_SWAPINTERVAL 0 -- 3.3 or 4.0. This turns off
      sychronizing to vertical blank.

      Run quake3 +set s_initsound 0 -- 3.3 or 4.0. Turn off sound.

      For 4.0 make sure nothing else is active on your screen. Making
      the X server active will eat cycles and cause contention from the
      hardware. The best way to do this is to run XFree86 and log in
      remotely to run quake. If you can't do that then leave just an
      Xterm running and start quake3 >& /dev/null. If you don't redirect
      output then you are waking up the Xterm and the Xserver to process
      that output.

If you do this, it would be interesting to see the numbers from a 3.3,
3.9.18 and possibly 3.9.16. My test rig is a PII-350. Q3A performance
remained the same for me. Gears got a good 10% faster between the 3.9
releases. It is possible the Q3A performance dropped, but I'm not sure
why. After we get benchmark numbers (be sure to include your hardware
description in there) we may be able to look at other profiling information.

						- |Daryll