Minor severity as the driver works as intended.
I find the low performance odd though. If the accelerator hardware
is slower than VESA for some operations, then surely the unaccelerated VESA
way could be used for those things. So it seems to me it doesn't
need to be slower in any cases.
My testcase is to play the game cuyo at levels 2 or 9. (package cuyo)
This is a tetris-like 2D game. It updated the screen 10 times
a second, and has a irritating failure mode: If it can't
finish drawing in 1/10s, then it refuse to take input
until the drawing queue clears. So it may break very noticeably
when there is too much animation going on.
Now, perhaps that isn't the best way to design a game, but
it shows a performance problem.
The ati driver (and the NV driver) easily falls into this trap when
there is lots of animation in the cuyo window. The game
becomes quite unplayable.
This simply doesn't happen with the VESA driver. CPU load stays
below 25% always.
The processor in this case is a 2.4GHz pentium-M,
lspci shows:
01:00.0 VGA compatible controller: ATI Technologies Inc Radeon RV100 QY
[Radeon 7000/VE]
Another machine with a pci radeon 9200 SE and a 1.8GHz opteron
shows the same problem.
A third machine with a 1.8Ghz core duo and a nvidia card has
this problem too, but of course that is a different package