furkan | has anybody noticed this text corruption in LibreOffice? https://www.dropbox.com/s/ypr3q8si3br2kze/libreoffice2.png?dl=0 | 03:22 |
---|---|---|
furkan | i wonder if it's X-related | 03:22 |
furkan | i'm guessing probably not, but figure it doesn't hurt to ask | 03:22 |
furkan | this is with 16.04 | 03:23 |
tjaalton | not here on intel | 04:13 |
alkisg | Hi, I'm trying to use glxgears in order to benchmark various window managers, compiz, metacity, xcompmgr etc. | 05:12 |
alkisg | But in one case I'm getting very weird results: http://paste.ubuntu.com/15682185/ | 05:12 |
alkisg | I.e. 472 FPS for one minute, then dang, 572 FPS the next minute | 05:13 |
alkisg | Nothing running on the client, no cron jobs or dmesg or xorg messages etc etc | 05:13 |
alkisg | It's even somewhat reproducible | 05:13 |
alkisg | ...what can possibly boost glxgears to have an additional 100 FPS after 1 minute of it running?! | 05:13 |
alkisg | (intel 855 graphics) | 05:17 |
alkisg | (ubuntu 16.04, plain xinit xterm) | 05:18 |
tjaalton | glxgears is not a benchmark | 05:24 |
alkisg | I've seen everyone saying that, and then offering no alternatives :) | 05:24 |
alkisg | But the numbers I care about are e.g. metacity --no-composite=500 fps, --composite=200 fps | 05:25 |
tjaalton | glmark2? | 05:25 |
alkisg | So it's some indicator. The performance difference in e.g. youtube is visible with the naked eye, I just want some hard numbers to get the developers interested in addressing them | 05:25 |
alkisg | Thanks, let me check that... | 05:25 |
tjaalton | also very good at hanging intel in the past.. | 05:26 |
tjaalton | skylake at least | 05:26 |
alkisg | Error: Glmark2 needs OpenGL(ES) version >= 2.0 to run (but version string is: '1.3 Mesa 11.2.0')! | 05:26 |
tjaalton | too bad | 05:29 |
* alkisg will try it with some newer client | 05:29 | |
alkisg | But any idea what could cause such a big performance difference in glxgears from one minute to the next, though? | 05:29 |
tjaalton | no | 05:29 |
alkisg | Thanks for the glmark2 hint, tjaalton :) | 05:29 |
tjaalton | sounds like it's running asynchronously, unlike on modern hw where it reports fps ~= vert refresh | 05:30 |
alkisg | I'm running it with vblank_mode=0 | 05:31 |
alkisg | Otherwise yeah I'm getting a constant 60 fps | 05:31 |
tjaalton | since you have obsolete hw just benchmark using gtkperf or such | 05:32 |
tjaalton | not that the wm matters there | 05:32 |
alkisg | It appears to matter a whole lot, metacity --composite has 1/3 of the performance of metacity --no-composite | 05:34 |
alkisg | ...in gtkperf, x11perf and glxgears | 05:35 |
alkisg | ...so I'm trying to convince them to put back --no-composite as the default (they changed it in 16.04) | 05:35 |
tjaalton | ok then | 05:36 |
alkisg | https://bugs.launchpad.net/ubuntu/+source/metacity/+bug/1566157 | 05:36 |
ubottu | Launchpad bug 1566157 in metacity (Ubuntu) "Metacity's compositing is too slow" [Undecided,New] | 05:36 |
tjaalton | I don't think anyone cares, really :) | 05:40 |
alkisg | Well, it makes the difference of "usable" vs "unusable" in thousands of pcs here... 5 fps in youtube vs 15 is a very big difference | 05:40 |
tjaalton | then change the default there | 05:41 |
alkisg | They replied that --no-composite is unsupported | 05:41 |
alkisg | So I'm evaluating different DEs currently, Mate seems to fare much better... | 05:41 |
alkisg | glmark2 score without wm: 2058, marco --no-composite: 1894, compiz : 1878, metacity --no-composite: 742, marco --composite: 708, metacity --composite: 441 | 06:45 |
=== darkxst_ is now known as darkxst | ||
tseliot | ricotz, mamarley: I've just uploaded 364.15 with KMS enabled, the initrd hook, and without the gfx blacklist. This is for 16.04, 15.10, 14.04 | 15:47 |
mamarley | tseliot: Yay, thanks! | 16:32 |
tseliot | :) | 17:35 |
* tseliot -> EOW | 17:35 | |
mamarley | Have a good weekend! | 17:38 |
=== darkness is now known as darkxst |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!