Anybody knows if Ubuntu 12.04 LTE will get new Intel Drivers in future (there where talk about new kernel versions for it), so one can test SNA?
Current driver does not support SNA, this is on old 945GME Netbook, so every bit helps...
Phoronix: SNA Delivers Huge Gains Over UXA For Intel "Gen5"
Chris Wilson's continued work on the SNA acceleration architecture is really beginning to pay off. In the latest SNA vs. UXA benchmarks for this 2D acceleration back-end to Intel's Linux graphics driver, SNA is delivering some heavy advantages...
http://www.phoronix.com/vr.php?view=MTMxMzM
Anybody knows if Ubuntu 12.04 LTE will get new Intel Drivers in future (there where talk about new kernel versions for it), so one can test SNA?
Current driver does not support SNA, this is on old 945GME Netbook, so every bit helps...
Haha Michael are you sure you tested the same CPU? Joking aside, these are awesome results.
AFAIK it will not by default, but metapackages to install the graphics stack of newer Ubuntu version will appear in the repos of precise after their release(they already have for Quantal). Also, newer precise ISOs will use those newer graphics stacks. Don't know how "clean" the installation of said metapackages is though(breakage, etc)..
Ok thanks, I really hope those metapackages works better than the x-swat ppa (2.19 intel driver-> no sna and mesa 9) + intel driver ppa (2.20.x) -> ironically breaks because of nouveau (on Atom N270 Netbook without any nvidia).
Going to try to purge x-swat ppa now and see if intel ppa alone works (perhaps it comes also with mesa 9 etc.?).
A bit more performance for cairo with sna would be nice.
Heureka, success!
Purging the x-swat ppa and only then activating ppa:glasen/intel-driver did the trick!
It comes with Mesa 9.0.3 and intel 2.21.3, X-swat had Mesa 9.0 and intel 2.19.
SNA is active (already had the needed entry in xorg.conf), doing some cairo-perf-trace benchmarks now...
But it's worse in GtkPerf?
Wow, really nice performance wins with SNA even with old GMA 945 chipset:
http://openbenchmarking.org/result/1...FO-CAIROPERF23
(334 seconds down to 118 seconds)
FYI, Compositing on was Mesa 9.0, Compositing off (not shown) was identical, all on XFCE.
Question is, do those GtkPerf Numbers have any real relevance in the real world? What does that really measure - never found any clue to that. Who cares if a simple Window can be drawn 1000 or 4000 times a second.
What counts imho: is the GUI slugish, is it drawn with less than 60 fps?
Somebody knows if GtkPerf can be any indicator to that?
Thats the reason I only benchmark cairo-perf-trace on a Netbook thats good only for browsing anyway, complex Opengl Benchs would be a waste anyway, 9 fps vs 4 fps is great, but still nothing worth in real life. Imho same with GtkPerf - just try if your GUI feels good/snappy.