I have two desktop machines with radeons - one, a phenom, has 8GB and an RS780L - that works fine, although I note that with the radeon driver Xorg.bin is taking up to 13% of (one) CPU (according to 'top') when it starts, but mostly less than 10%, often only 2-3% when the box is idle and the CPU frequency has dropped off, but if I change to the modesetting driver the usage is normally under 1% (I did once see 3% after I closed firefox).
But the other machine is an A4 with the Aruba graphics, and it was here that I first noticed the problem. This box is underpowered (its main use is for when I have to run windows (tax, one music store, satnav updates), although I have some linux systems there. My desktop of choice is icewm (lean, but with a network window to remind me that a backup is running, and a CPU window). Some time last year on the A4, I noticed that icewm's CPU window showed a lot of red (possibly, that is memory pressure) in its lower part after Xorg started. At the time, I assumed that irregular fcron jobs such as updatedb were running (the individual linux systems do not get booted often, so running a weekly job would not be surprising). But today I wanted to take a look at how the modesetting driver compared on the A4 (this is the 1.16.3 server, with released drivers. I already knew that the modesetting driver reduced the apparent Xorg.bin CPU usage on the phenom. The A4 only has 4GB (less 256MB for graphics), and Xorg.bin is not using a lot - only 0.8% of memory - and the box was not swapping. Testing was with linux-3.19.0-rc5. After running startx, the CPU% for Xorg.bin was initially around 85% (icewm, two urxvt terminals, one running top, and me typing notes in the other). Left it, when I came back and moused to stop the screensaver it was again using 85-90% (this is presumably with the processor frequency at minimum), but after a couple of minutes it had fallen below 30%. I then did some minor compiles, including the modesetting driver, and watched the Xorg.bin CPU% running between 12% and 20%. During the compiles, I noticed from my power meter that the total system power consumption (monitor, net switch, kvm switch, this box) went up for a while - presumably, both cores running at full speed - and the Xorg.bin CPU% varied between 9% and 20%. Left it again, came back, ran some screensavers from xscreensaver-demo (Morph3D, MoebiusGears, GLHanoi fullscreen, CPU% seemed to be 2.0 to 2.7% immediately after each of them). When I closed the small window from xscreensaver-demo the CPU% again shot up to 15 to 20% briefly. At that point I closed X, then reran startx. As soon as I could start 'top', I saw that the CPU% was around 45%, but after a few seconds it dropped into a range between 30% and 42% : much better than the first time, but not wonderful. I then renamed the ati and radeon drivers so that they would not be found, and restarted X with the modesetting driver. Top showed Xorg.bin at 6.4% on the first reading, then 4%, but after that 2.0 to 2.3% and the system was idling with at 103-104W instead of the (initial) 125W and then 113-114W. With xscreensaver-demo (i.e. just a *small* window showing a random saver), Xorg.bin CPU% was typically 5%. Unfortunately, I could not run any of the interesting screensavers mentioned above, nor glxgears. That is my problem, if I care about it enough. FWIW it's a Mesa/LLVM problem - I had something similar in qemu which was solved by rebuilding Mesa. I've just started top again (modesetting driver) - Xorg.bin again 6.3% on first sample, but then 1.0 to 1.3%. Anyway, I'd rather have a lower power usage, or more CPU devoted to what I want to run, but it would be _nice_ to have the correct driver using less CPU ;) Any suggestions, please ? ĸen -- Nanny Ogg usually went to bed early. After all, she was an old lady. Sometimes she went to bed as early as 6 a.m. _______________________________________________ xorg-driver-ati mailing list [email protected] http://lists.x.org/mailman/listinfo/xorg-driver-ati
