[01:44] <Willis> I was playing around, and no matter what settings (playing with snd_streaming and r_font_hinting) on the second run, performance was fine, meaning even if it is using more cpu, it's not enough that it's cpu bound
[01:54] <LordHavoc> Willis: I've been really tempted to make DP somehow do an automatic twice-run timedemo, reporting time for only the second
[01:55] <Diablo-D3> why?
[01:55] <LordHavoc> because the engine uploads and compiles shaders on first use
[01:55] <LordHavoc> it also uploads lightmaps in q1bsp on first use
vid_restart causes the shaders and lightmaps to be flushed from this cache, hence if you do vid_restart between timedemos, you get the same result as if you restart.
Would it be worth using Phoromatic?
http://www.phoromatic.com/
That way an automated build of each DP engine code revision could be benched via the Nexuiz benchmark in PTS. We'd then have an official measure for performance rather than occasional claims for drops in performance which we can't necessarily depend upon.
This same approach is being done by Phoronix to track performance in the kernel Git repository which should bring payoffs for the whole Linux community.