|
|
#1 | |
|
Registered User
Join Date: Jun 2006
Posts: 12
|
I have a new PC with AMD X2 4850e and nVIDIA 8600 GT. After installation of openSUSE 11.1 (x86_64) I added the nVIDIA repository and installed the drivers (packages x11-video-nvidiaG02-177.82-1.1
nvidia-gfxG02-kmp-default-177.82_2.6.27.7_9.1-1.1 ) and configured X11 to use them. Everything worked like a charm. To my surprise, I noticed a very high CPU load when the PC was supposed to idle (eg. during night) (monitored by munin/sysstat). "High CPU load" reads: - CPU frequency rises to max (2.5GHz) - load average (as reported by uptime) > 1.2 - According to /proc/stat one of the 2 CPU cores is 100% busy, ~50% in system/kernel context. The other core reports ~90% idle To compare: such a system load appears only during video/audio processing. A usual full interactive KDE3.5 session (full KDE environment with bells and whistles, webbrowsing, email, ~2 xterm sessions) leads to - CPU frequency min (1.0GHz), small peaks when eg. starting OpenOffice - load average ~0.4 - Both cores ~75% idle Since the whole idea of the new PC was a lower power consumption (around 30W instead of >80W) I tried to figure out the reason. In short, it appears to me that the nVIDIA drivers in the openSUSE repository (in the packages mentioned above) are somewhat crippled: - The CPU load was caused by xscreensaver running OpenGL screensavers during the night. - This is not a bug in xscreensaver, since it usually causes very low CPU load (veryfied on several compareable PCs) - The problem vanished after installing the driver from NVIDIA-Linux-x86_64-177.82-pkg2.run: 1) OpenGL applications cause fewer CPU load in general 2) Now xscreensaver detects that the screen is powered down (DPMS) and logs: xscreensaver: 12:24:54: 0: X says monitor has powered down; not launching a hack. CPU load during nighttimes are now: - CPU frequency min (1.0GHz) - load average <0.15 - Both cores >85% idle Same driver version, totally different performance ![]() Sorry, but IMO it would be a better solution to _not_ offer a repository at all under these circumstances (or brand them as crippled). The benefit of very easy installation of the drivers is revoked by their lack of capability. Most people (especially Linux newbies) will suffer the same as me, but won't be able to figure out the reason. They will blame Linux instead of nVIDIA since they installed the newest stable driver from the manufacturer. BTW the attached log was made after installing the drivers from the .run file EDIT: by mistake, I installed 177.80 I updated to 177.82 and AFAICS the results are the same: the is definitly a huge performance difference between the drivers form the repo vs. self-compiled. |
|
|
|
|
|
|
#2 | |
|
Registered User
Join Date: Aug 2003
Posts: 1,026
|
This seems to be a quite complicated issue.
There are many factors in this bug, and now I read for the first time that the driver from the repository is involved in it as well. It is not really caused by the screensavers, the problem occurs as well when the xscreensaver package is uninstalled and the screensaver disabled. It seems related to the "powerdevil" component that gets run by the "kded4" process. When that process is killed (possible only when no KDE4 programs are running) the problem vanishes. But others report that it occurs in GNOME as well... Hopefully nvidia will look into the Nvidia-repository and also make available the new beta driver as an optional package in this repository. I like to use the driver from the repository but right now there seems to be a problem and I cannot try the beta driver either. |
|
|
|
|
|
|
#3 | ||||
|
Registered User
Join Date: Jun 2006
Posts: 12
|
Quote:
Think about how many ppl gave openSUSE 11.1 a try over christmas and are now disappointed. Eg. I read several threads where ppl are blaming the new kernel process scheduler CFS for a OpenGL performance penalty with openSUSE 11.1 I'd say they are having the same problem I had. Quote:
Quote:
Anyway, this is target practice in the dark since the nVIDIA drivers are closed source and there is no way to definitly tell from the outside which component is causing which effect. Quote:
I guess they are hearing this a lot :mrgreen: |
||||
|
|
|
![]() |
| Thread Tools | |
|
|
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| Maintain Your Privacy by Manually Accepting and Rejecting "Cookies" (nV News) | MikeC | Open Forum | 2 | 02-02-13 07:15 PM |