Closed totaam closed 10 years ago
What you are describing is a driver bug. Repainting a window should not be able to lock your session in this way.
(this ticket is somewhat similar to #353)
r5353 slows things down a bit, but there is no way we can fix buggy drivers.. So I am closing this as notabug, feel free to re-open if this can be reproduced with other, non-buggy, driver + card combinations.
Replying to [comment:1 totaam]:
(this ticket is somewhat similar to #353)
Somewhat similar except that I've never had a problem with high CPU usage. If I'm not mistaken #353 was reported even before Xpra got OpenGL support... This is about high GPU usage...
What you are describing is a driver bug. Repainting a window should not be able to lock your session in this way.
r5353 slows things down a bit, but there is no way we can fix buggy drivers.. So I am closing this as notabug, feel free to re-open if this can be reproduced with other, non-buggy, driver + card combinations.
I'm not convinced that it is a driver bug although it might be a bug. It could be inefficient driver or GPU is just not powerful enough. One spinner do not cripple affected system but multiple spinners do. Let's consider hypothetical (but not uncommon) case when 10 (or more) applications are running in session. How many simultaneous spinners you would consider reasonable to run to utilise nearly 100% GPU while normal OpenGL acceleration of all those windows hardly use 50%?
Even re-painting at 10fps, painting simple shapes like this, you should be able to paint hundreds of windows without noticing the extra load. Even on an outdated GPU, say a 1990s GPU. Even those were able to paint hundreds of textures per second (and not just plain rectangles).
I've just tested by running the client with 10 windows and:
XPRA_FAKE_BROKEN_CONNECTION=1 xpra attach ...
xpra is at 1% cpu usage, Xorg at about 10% with a compositing manager. The GPU is idle, on its lowest power setting.
Makes perfect sense, thanks. Buggy driver (or card) it is...
Please see #818#comment:6, I think this is probably fixed by this backport.
Issue migrated from trac ticket # 646
component: client | priority: major | resolution: invalid
2014-08-20 06:27:04: onlyjob created the issue