unstable graphic device (high CPU consumption and crashes with vid-streams)

[michaa7]

I bought a new graphic adapter

$ inxi -G
Graphics:  Card: NVIDIA G84 [GeForce 8600 GT]
           Display Server: X.Org 1.16.0 driver: nvidia Resolution: 1680x1050@59.88hz
           GLX Renderer: GeForce 8600 GT/PCIe/SSE2/3DNOW! GLX Version: 3.3.0 NVIDIA 340.32

which presents a much better video performance on my old and week system

$ inxi -C
CPU:       Single core AMD Sempron 2600+ (-UP-) cache: 128 KB clocked at 1607 MHz

I now can use DVI instead of VGA which is the only output my onboard nvidia 6100 offered.

Now the drawback: Watching TV streams sooner or later lets the system crash completely. No freeze, but a kernel panic or the like. First the video freezes but immediately after I have two blinking diodes on my keyboard and I have to hard reset my system. This happens repeatedly each 15 to 20 minutes (while watching a video stream. Videos from hd are fine).

I assume that the high CPU consumption which with this new card and driver (nvidia 340) is continuously near to 100% might be the cause for this crash. In other words: The CPU gets too hot. Or are there other explanations?

Anyway, how does a better card with built in hw acceleration need more CPU power than a weeker card without hw acceleration?

What should I test?  The problem is, I have no clue which logs could be helpfull. As I have to hard resett the system I doubt there is any usefull information. I will try to run htop from a ssh connected laptop to observe the CPU stress. As far as I can tell (but not sure) there is no sensor for CPU temperature.

As it is currently, the fun with this card is marred (as far as streams are concerned).

So what do I do (other then sending back the card)?

[mxmehl]

Do I understand you correctly that these problems were not existent on the system before the new graphic card?
And I assume that you mean flash streams when speaking of TV streams (e.g. wilmaa.com).

In an old system I had similar problems caused by flash streams. The CPU was on 100% usage the whole time, fortunately I didn't have any kernel panics or hard reboots. I solved this either by using gnash or by reinstalling flashplugin-nonfree (Adobe).

Wait, one more question: You speak about hw acceleration. AFAIK the adobe plugin doesn't support it. But well, please provide more information about what streams you are talking about.

[michaa7]

Do I understand you correctly that these problems were not existent on the system before the new graphic card?

Yes, those problems were *not* existent with the old, weeker onboard graphic. But the quality is better with the newone, as long as it doesn't crash ;-) .

And I assume that you mean flash streams when speaking of TV streams (e.g. wilmaa.com).

Yes. It happens with all flash streaming content like (you can't access from outside Germany!!! unless you use a German proxy) http://www.zdf.de/ZDFmediathek/beitrag/live/1822600/Das-ZDF-im-Livestream?bc=lve;lv0&flash=off

In an old system I had similar problems caused by flash streams. The CPU was on 100% usage the whole time, fortunately I didn't have any kernel panics or hard reboots. I solved this either by using gnash or by reinstalling flashplugin-nonfree (Adobe).

I'give it (reinstall) a try.

Wait, one more question: You speak about hw acceleration. AFAIK the adobe plugin doesn't support it. But well, please provide more information about what streams you are talking about.

I disabled hw-acceleration in flashplugin.

As for the streams, I am trying now thisone. It is long enough to trigger the bug or whatever:
http://www.youtube.com/watch?v=rz5eeA_BxrQ

I'll report whether on not it *successfully* crashed the system.

EDIT: it didn't crash until now. But I have reinstalled flashplugin-nonfree. I'll see whether it make any difference on other streams which used to crash.

But as the CPU usage was quite continuously at 100% (and it "drops" to 99.x sometimes), it didn't make me feel comfy. Having such CPU usage can't be a solution even if the system does noch crash.

EDIT2:
I now "successfully" crashed the system with a youtube flash stream. With youtube you can adjust the quality. With the lowest, I only have about 60-70% CPU workload, on the second highest (420p) it now did run quite an hour, but all the time @ 100%.

So this clearly is flash related. Unfortunatly not all streaming services allow for adjusting quality. And the workload *is* insame.

I will try the nvidia legacy 304 driver to see whether or not the devoloppment did go versus better or worse.

[bluelupo]

Hi michaa7,
have you ever CPU temperatures measured over a longer period of time and under stress. I can not imagine, however, that too high a temperature of the CPU has a system crash resulted.

[michaa7]

I don't know whether my board offers this. I know lm-sensors. But there is no "man lm-sensors". I once used it, but I forgot how to use it. Any indications?
EDIT:
Ok, found it, lm-sensors + psensor
/EDIT

And maybe it is not the temperatur. But running at 100% for an hour may lead to a situation where some little timing problem may cause the CPU to get out of step. that's how I assume it happens.

I now figured out that setting the flash quality to 320p prevents the CPU workload to reach 100%. It is still high, between 80% and the highes 99%, but does not really reach 100% or even being there for a long time.

[bluelupo]

Hi michaa7,
you can install the sensord package for more infomation.

Code Select Expand


Package: sensord
Source: lm-sensors
Version: 1:3.3.5-2
Installed-Size: 69
Maintainer: Aurelien Jarno <aurel32@debian.org>
Architecture: amd64
Depends: lm-sensors, lsb-base (>= 3.2-13), libc6 (>= 2.7), librrd4 (>= 1.3.0), libsensors4 (>= 1:3.1.1)
Suggests: rrdtool
Description-de: Log-Dämon für Hardware-Sensor-Informationen
Lm-sensors ist ein Hardware-Zustands-Beobachtungspaket für Linux. Es erlaubt
Ihnen, Informationen von Temperatur-, Spannungs- und Lüfter-Sensoren auszulesen.
Es arbeitet mit den meisten neueren Systemen zusammen.
.
Dieses Paket enthält einen Dämon, der Hardware-Zustandsstati im Systemlog
verzeichnet mit zusätzlichen Warnungen zu möglichen Systemproblemen.
[...]


[michaa7]

ok, habe ich nun zusätzlich installiert. Mir ist nicht klar ob das schon installierte psensor ein eigenes log schreibt, aber das syslog sollte es ja wohl auch tun.

Ich habe es jetzt mal beobachtet. Die Normal IDLE temperatur liegt wohl bei 42°C, normales browsen führt zu ~46°C, mit Videostreams steigt dann  die Temperatur auf 54°C/129°F .

Ich werde nach dem nächsten crash mal schauen was mir

Code Select Expand

# cat /var/log/syslog | grep sensord

sagt.

Danke soweit.

[towo]

Du kannst auch folgendes probieren:

Code Select Expand

mkdir -p /etc/adobe;echo EnableLinuxHWVideoDecode=1 > /etc/adobe/mms.cfg

dann läuft flash über vdpau, allerdings hatte das bei mir den Effekt, das Youtube wunderbar ging, andere Seiten zeigten dann aber nicht mehr viel Sinnvolles an.
Einen Test ist es aber allemal wert.
Aber hier erzeugt Flash nicht solch eine hohe CPU-Last, und ich habe die Beschleunigung auch deaktiviert.
Allerdings ist ein Single core AMD Sempron 2600+ auch keine Rakete.

[michaa7]

Unerklärlicherweise hat das den zeitpunkt bis zum crash nur beschleunigt. Ich habe das mit

Code Select Expand

mv /etc/adobe/mms.cfg /etc/adobe/mms.cfg.savenun erstmal revidiert.

Bin am überlegen die vorhandene CPU Sempron64 2600+ gegen eine Athlon64 3200+ (die neben höherm tackt den 4fache L2 cache und leicht niedrigeren Verbrauch hat. Wer sowas noch zu Hause rumfleigen hat, es geht um Sockel 754) zu tauschen. Was sagen die fachleute, ist da eine signifikante leistungsverbesserung zu erwarten die den umbau und 20€ rechtfertigen?

[der_bud]

...ist da eine signifikante leistungsverbesserung zu erwarten die den umbau und 20€ rechtfertigen?

Nein, definitiv nicht, der Unterschied zwischen den beiden Prozis ist nicht so signifikant. Eine deutlich billigere Lösung siehe -> PM  .

[michaa7]

...ist da eine signifikante leistungsverbesserung zu erwarten die den umbau und 20€ rechtfertigen?

Nein, definitiv nicht, der Unterschied zwischen den beiden Prozis ist nicht so signifikant. Eine deutlich billigere Lösung siehe -> PM  .

Eben erst nach antwort auf PM hier deine einschätzung gelesen. Das bestätigt meine zweifel im grunde. Da kann ich die 20€ auch woanders hinlenken ;-) .