CPU speed scaling stability problems

Message

hvengel · Post by **hvengel** » Tue Dec 20, 2005 11:03 pm

Starting with kernel 2.6.12 I starting having random stability issues. Mostly it manifested as firefox crashing but I was also noticing that the system would occationally hang when running an emerge. Since it was a rare occurance I did not spent too much time trying to find the source of the problem.

Then last week I upgraded my CPU to an X2 (from a 3500 Winchester) and my system became very unstable with gcc segfaulting and system hangs. I also have Windows on the same machine I noticed that when I installed the X2 capable CPU driver Windows was also very unstable. After removing the CPU driver on Windows it became stable. So I booted from the Gentoo install disk and chrooted into my system and rebult the kernel without frec. scaling and now my system is rock stable. Clearly the best it has been since 2.6.11.

Has anyone else noticed anything like this? And what should I do to trouble shoot the problem so that I can get freq. scaling working again?

mudrii · Post by **mudrii** » Wed Dec 21, 2005 7:02 am

why you do not try new kernel 2.6.14 ? and rebuild the system with new hardware and it should work thought.

HermanR · Post by **HermanR** » Wed Dec 21, 2005 1:45 pm

I had problems related to frequency scaling too. My new AMD64 3200+ computer crashed a lot (and bigtime: forcing me to use the power button to get things going again), but ONLY when it was (next to) idle. For some time, I didn't know what to do with that, until it occurred to me that it could have been caused by the CPU operating at a lower frequency (ondemand governor). I disabled the frequency scaling stuff and since then my computer has been 100% stable. I really liked frequency scaling and I had it working perfectly, but I prefer a stable system that I can leave on during the night and can be sure to be still running when I return to it in the morning!

deepspace9 · Post by **deepspace9** » Wed Dec 21, 2005 9:25 pm

Same here. I got probs with frequency scaling too (2.6.14-gentoo-r2). Shut it down, and now everything runs just fine again...

Oh well, my system reamains cool withou it, too

HermanR · Post by **HermanR** » Wed Dec 21, 2005 10:04 pm

Well, no watercooling here, but my CPU never reaches even 40 degrees Celsius, thanks to the Zalman cooler. My previous Athlon at 900 MHz topped 75 degrees!!

deepspace9 · Post by **deepspace9** » Wed Dec 21, 2005 10:07 pm

HermanR wrote:Well, no watercooling here, but my CPU never reaches even 40 degrees Celsius, thanks to the Zalman cooler. My previous Athlon at 900 MHz topped 75 degrees!!

Mine does 45 max, but I have an X2

HermanR · Post by **HermanR** » Thu Dec 22, 2005 11:00 am

deepspace9 wrote:Mine does 45 max, but I have an X2

Yeah, yeah, rub it in...

And that from a fellow countryman!

hvengel · Post by **hvengel** » Thu Dec 22, 2005 9:48 pm

I am using 2.6.14-r2 so the speed scaling problems are not fixed in the latest version. I was running this version with my 3500 Winchester before I upgraded to X2. I have rebuild the kernel since teh CPU upgrade.

Well it is interesting that others are having the same problem. Too bad no one seems to know how to fix it. My systems is also fairly cool. The CPU will hit 45C when running a large emerge and it idles at about 35C in a 19C room. The hottest I have seen it get is when running two copies of cpuburn when it peaked at about 53C after about 20 minutes. But beyond the fact that idle temps would be considerably lower with CPU speed scaling the machine would also use less power. Less power consumption means a cooler power supply which results in lower noise levels. So I would really like to get it working again.

aidy · Post by **aidy** » Mon Dec 26, 2005 11:35 pm

HermanR wrote:
deepspace9 wrote:Mine does 45 max, but I have an X2
Yeah, yeah, rub it in... And that from a fellow countryman!

Remeber what you are

Post by **pacho2** » Tue Dec 27, 2005 7:44 pm

Are you overclocking your CPUs?

HermanR · Post by **HermanR** » Tue Dec 27, 2005 11:07 pm

pacho2 wrote:Are you overclocking your CPUs?

Short answer: no!
Long answer: no, I don't!

janiskr · Post by **janiskr** » Wed Dec 28, 2005 12:58 am

try to upgrade BIOS, atleast for DFI mobos it helps if scaling does not work at all

Post by **pacho2** » Wed Dec 28, 2005 12:41 pm

HermanR wrote:
pacho2 wrote:Are you overclocking your CPUs?
Short answer: no!
Long answer: no, I don't!

Try to update your BIOS. I have an Athlon 3200+ (2Ghz - overclocked to 2.31) and I have no problems with powernow. My mother board is an Abit AV8.

hvengel · Post by **hvengel** » Thu Dec 29, 2005 12:33 am

I have updated my BIOS to the latest. Didn't change a thing as far as CPU speed scaling stability.

I am not over clocked, my memory is running at default speed and everything is set to more or less default settings in the BIOS. memtest86 says everything is OK even after hours of tests.

I had a 3500 Winchester in this same machine for a while and it was rock stable with CPU speed scaling turned on until 2.6.12 and then I started having some signs of stablility problems. Specifically firefox would crash every now and then. But it was not often enough to be a real concern. I thought that perhaps v 1.06, which came out about the same time as kernel 2.6.12, had a problem. When I dropped the X2 in it became totally unstable until I shut down CPU speed scaling. Now it is totally stable and even firefox is now totally stable.

On the AMD X2 drivers page AMD has code for Linux (version 1.2.2...) and if you download that and open it up and have a look it says in the README (I think that where I saw it) that this code was included in the official kernel starting with 2.6.12. Wow what a coincidence the CPU speed scaling code changed at the same time that my machine started to have stability problems that were fixed by removing the CPU speed scaling code.

It appears that others have also seen some of the same problems. So I guess that we will have to wait for AMD to fix the code.

mudrii · Post by **mudrii** » Thu Dec 29, 2005 1:08 am

I use ASUS A8N SLI Premium with AMD X2 3800+ with Kernel 2.6.14 with CPU speed scaling turned on and every thing just works with 0 problem rock stable and with hi speed.

wantilles · Post by **wantilles** » Thu Dec 29, 2005 12:54 pm

2.6.14 kernels have finally solved all the remaining micro-glitches (real-time clock etc.) with the X2 CPUs.

Cool & Quiet and thermal throttling have been running flawlessly from 2.6.12.

You should upgrade your kernel to gentoo-sources-2.6.14-r5.

Writing this from Gentoo amd64 running on a Toledo 4400+ X2.

HermanR · Post by **HermanR** » Sat Dec 31, 2005 3:58 pm

Okay, I've re-enabled cpu frequency management in my kernel (gentoo-sources 2.6.14-r5). My system is running now at 1000 MHz (maximum 2000 MHz) with the ondemand governor, just as I had it working in the beginning.

Now let's see if everything crashes again, or whether the problems were kernel-related and have been fixed in the meanwhile. So far (3 hours uptime), so good.

But I'm keeping a cpufreq-less kernel in my grub config, just to be on the safe side.

EDIT: After 11 hours, my system froze again. So I'll stay away from cpu frequency management, since this solves my problem without any doubt, whether the problem is in the kernel or in my hardware. Happy New Year everyone, btw.

hvengel · Post by **hvengel** » Sun Jan 01, 2006 8:30 pm

I was running 2.6.14-r2 when I upgraded to X2 and started having problems with CPU speed scaling. I also upgraded to 2.6.14-r5 as soon as it became available and it did not fix the problem. I have done some streamlining of my kernel config since that time and I will give r5 another try later on.

halfgaar · Post by **halfgaar** » Thu Feb 16, 2006 11:39 pm

Allow me to dig-up this thread.

I just finished building my X2 3800+ system and was experimenting with cool'n'quiet. I haven't got any stability problems yet, but I have this nasty thing in Unreal Tournament: it runs much too fast with c-n-q enabled. I guess it determines speed based on some timer, and then when the clockspeed increases, the game runs way too fast... Is there a fix for that? I would imagine other apps also react badly to this.

My kernel is gentoo-sources 2.6.14-r2.

Also, I am somewhat in doubt about all those temperatures I'm seeing here. My system is watercooled, and the CPU gets 42 C when idle without c-n-q (about 40 with c-n-q), and 52 when both cores are at 100%. These are the temperatures that lm-sensors give me. But I find it rather strange that people with stock-coolers get 35 C when idle... (could have seen that reading in another thread, I'm not sure). I don't have much faith in those readings...

btlee · Post by **btlee** » Fri Feb 17, 2006 2:32 am

I have the same experience.
In my case, always freezing occurred during downscaling.
Upscaling made no freezing.
Just my guess is that bios or kernel driver for cpu scaling could not support specifically for voltage control.
So, now i don't use CnQ.

r.abbott · Post by **r.abbott** » Thu Mar 02, 2006 11:55 pm

Question from a lowly non-CS: how do you guys see your CPU temperature? I assume there's some program or /proc entry, but I haven't been able to find it.

halfgaar · Post by **halfgaar** » Fri Mar 03, 2006 12:01 am

You have to install lm_sensors, run sensors-detect to find out what kernel module you need and enable it in the kernel. There are docs that describe this in detail, on the lm_sensors site for example.

r.abbott · Post by **r.abbott** » Fri Mar 03, 2006 3:51 am

Very nice, thanks. I had to patch the kernel to make lm_sensors work with my (less-than-ideal) ATI chipset, but it does work.

frank1980 · Post by **frank1980** » Fri Mar 03, 2006 5:16 am

Just BIOS support "cool 'n' quiet" and the AMD cpu would change its frequancy automatically without any setup under KDE enviroment?

halfgaar · Post by **halfgaar** » Fri Mar 03, 2006 12:08 pm

There is some configuration to be done. You can let the kernel determine the frequency, or a userland program. I believe the latter is a thing of the past now.

See "http://gentoo-wiki.com/HOWTO_PowerNow!" (the exclamaition mark ruins phpbb's url-tag). It's up to you to set either the ondemand or conservative governor. Ondemand is more logical to me, because conservative waits too long with switching the frequency. It's more suited to laptop use, the kernel says so as well.

But, I disabled cool'n'quiet, because I do notice slowdown in repsonsiveness, and I suspect the difference in powerusage is not big. There is only a 2 degree C difference in temperature between idle without c'n'q, and with c'n'q. When the CPU is stressed, the temp raised about 10-12 degrees, so even without c'n'q, it's quite low on power. And, unreal tournament doesn't respons well to CPU frequency changing...