Hello,
some similar problems have already been reported. However, this looks
a bit different. With a large number of recent kernels (2.6.4 and 2.6.5,
mm-series and rc-series) the temperature of my CPUs is *much* higher
than with 2.6.3-mm4 (and previous). My board is a Tyan Tiger MPX with
2x Athlon 2000+ MP. The temperature is about 38-45 Celsius (on low load)
with 2.6.3-mm4. Later kernels show 48-55 Celsius. I'm using amd76x_pm
for power-saving (both linux-2.6-amd76x_pm-20031109.patch and
patch-2.6.4-amd76x_pm). At the moment this is not a big problem as
2.6.3-mm4 works perfect for me. However, there might be future changes
that make me wanting a newer kernel...
I did some tests with nmi_watchdog and the proprietary Nvidia graphics
driver and didn't find any impact. There's no significant difference
using no nmi_watchdog, nmi_watchdog=1 (only 2.6.3-mm4) or
nmi_watchdog=2. I also did some tests without an X server running:
temperatures go down to the lower end but the difference of 10 Kelvin
between 2.6.3-mm4 and 2.6.5 persists. I didn't try any of the acpi
kernel-parameters.
See attached config for 2.6.5 for details.
Any ideas?
-jo
--
-rw-r--r-- 1 jo users 80 2004-04-05 21:34 /home/jo/.signature
On Tuesday 06 April 2004 15:36, Joerg Sommrey wrote:
>Hello,
>
>some similar problems have already been reported. However, this
> looks a bit different. With a large number of recent kernels (2.6.4
> and 2.6.5, mm-series and rc-series) the temperature of my CPUs is
> *much* higher than with 2.6.3-mm4 (and previous). My board is a
> Tyan Tiger MPX with 2x Athlon 2000+ MP. The temperature is about
> 38-45 Celsius (on low load) with 2.6.3-mm4. Later kernels show
> 48-55 Celsius. I'm using amd76x_pm for power-saving (both
> linux-2.6-amd76x_pm-20031109.patch and patch-2.6.4-amd76x_pm). At
> the moment this is not a big problem as 2.6.3-mm4 works perfect for
> me. However, there might be future changes that make me wanting a
> newer kernel...
>
>I did some tests with nmi_watchdog and the proprietary Nvidia
> graphics driver and didn't find any impact. There's no significant
> difference using no nmi_watchdog, nmi_watchdog=1 (only 2.6.3-mm4)
> or
>nmi_watchdog=2. I also did some tests without an X server running:
>temperatures go down to the lower end but the difference of 10
> Kelvin between 2.6.3-mm4 and 2.6.5 persists. I didn't try any of
> the acpi kernel-parameters.
>
>See attached config for 2.6.5 for details.
>
>Any ideas?
Recently I've had to change the multiplier in gkrellm to reduce it by
a div/10 to get the correct reading. I've no idea how to make
sensors see the new, multiplied by 10 readings correctly, I cannot
make it run with later 2.6 kernels.
For gkrellm-2.1.28, the .28 version is needed by later kernels anyway,
use the default, it apparently already corrects for this with its
1.000 multiplier. 2.1.24, when it worked, had to have a multiplier
of 0.1000.
But join the 70C club, that AMD athlon keeps itself at a medium simmer
full time. Mine has been running 67-72C for 3 years now. Strangly,
shutting down setiathome doesn't cool it by more than a couple
degrees C. And, its got a $50 all copper Glaciator cooler on it,
heavy heavy heavy.
--
Cheers, Gene
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
99.22% setiathome rank, not too shabby for a WV hillbilly
Yahoo.com attornies please note, additions to this message
by Gene Heskett are:
Copyright 2004 by Maurice Eugene Heskett, all rights reserved.
On Tue, Apr 06, 2004 at 04:26:37PM -0400, Gene Heskett wrote:
>
> But join the 70C club, that AMD athlon keeps itself at a medium simmer
> full time. Mine has been running 67-72C for 3 years now. Strangly,
> shutting down setiathome doesn't cool it by more than a couple
> degrees C. And, its got a $50 all copper Glaciator cooler on it,
> heavy heavy heavy.
That's not quite my point. I am not afraid of running my athlons at
70C. I just don't want to. With Debian Woody they ran at <40C, which
is impressing IMHO. An upgrade to Sarge raised the temp for about 5K,
which is still very cool. This temperature didn't change when I
upgraded to an early 2.6 kernel. Just after 2.6.3-mm4 there was this
jump for 10K that I just do not understand. It doesn't hurt the athlons
but seems unnecessary to me.
-jo
--
-rw-r--r-- 1 jo users 80 2004-04-06 18:59 /home/jo/.signature
On Tuesday 06 April 2004 16:45, Joerg Sommrey wrote:
>On Tue, Apr 06, 2004 at 04:26:37PM -0400, Gene Heskett wrote:
>> But join the 70C club, that AMD athlon keeps itself at a medium
>> simmer full time. Mine has been running 67-72C for 3 years now.
>> Strangly, shutting down setiathome doesn't cool it by more than a
>> couple degrees C. And, its got a $50 all copper Glaciator cooler
>> on it, heavy heavy heavy.
>
>That's not quite my point. I am not afraid of running my athlons at
>70C. I just don't want to. With Debian Woody they ran at <40C,
> which is impressing IMHO. An upgrade to Sarge raised the temp for
> about 5K, which is still very cool. This temperature didn't change
> when I upgraded to an early 2.6 kernel. Just after 2.6.3-mm4 there
> was this jump for 10K that I just do not understand. It doesn't
> hurt the athlons but seems unnecessary to me.
>
>-jo
40C? Shut down for an hour to cool, I've never seen the post on my
board show less than 63C by the time it gets to that part of the
bios. I'm running a 1400DX at 1400mhz, so the bios thinks its a
1600DX, and I've got vcore set down to 1.65 volts which helps a bit.
Actually, the athlons seem to have a builtin shutdown at 75C, I've hit
that once or 3 times when the air under the desk was trapped worse
than usual. Makes for downright ugly reboots...
--
Cheers, Gene
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
99.22% setiathome rank, not too shabby for a WV hillbilly
Yahoo.com attornies please note, additions to this message
by Gene Heskett are:
Copyright 2004 by Maurice Eugene Heskett, all rights reserved.
* Joerg Sommrey <[email protected]>:
> On Tue, Apr 06, 2004 at 04:26:37PM -0400, Gene Heskett wrote:
>>
>> But join the 70C club, that AMD athlon keeps itself at a medium simmer
>> full time. Mine has been running 67-72C for 3 years now. Strangly,
>> shutting down setiathome doesn't cool it by more than a couple
>> degrees C. And, its got a $50 all copper Glaciator cooler on it,
>> heavy heavy heavy.
>
> That's not quite my point. I am not afraid of running my athlons at
> 70C. I just don't want to. With Debian Woody they ran at <40C, which
> is impressing IMHO. An upgrade to Sarge raised the temp for about 5K,
> which is still very cool. This temperature didn't change when I
> upgraded to an early 2.6 kernel. Just after 2.6.3-mm4 there was this
> jump for 10K that I just do not understand. It doesn't hurt the athlons
> but seems unnecessary to me.
At least on A7M266-D lmsensors read thermal sensors very wrong. I
haven't got time to contact devs with that, but I do know for sure that
amd76x_pm really does make cooling calls, even in 2.6.5-rc3-mm3
(There should be /sys/devices/pci0000\:00/0000\:00\:00.0/C2_cnt file,
which tells how many times has amd76x_pm really made the disconnection call).
One issue is that from some kernel version amd76x_pm's idle() is called
upto 3.5x times more often when there's some audio activity. So in
effect number of calls to default_idle() jumps from 1100Hz to 3800Hz.
(this is reproducible with 'rhytmbox' -application, but not with xmms.
AFAIK my xmms uses OSS emulation and rhytmbox is native alsa.)
Ahem. Could you actually try:
echo 3 > /sys/devices/pci0000\:00/0000\:00\:00.0/lazy_idle
This could help gaining 5-8?C. HZ changed from 100 to 1000 in 2.6, so
amd76x_pm old default doesn't apply overly well here.
There's some funniness going on with this tunable. It doesn't really
affect how many times/second we call amd76x_pm.idle(), but rather how
easily we go into sleep (no sleep if both CPU's aren't idle).
With lazy_idle at 3 I get bad distortions with bttv card. with 3000 they
disappear, but so does the thermal throttling :)
(Sorry for lack of coherence right now)
--
Psi -- <http://www.iki.fi/pasi.savolainen>
On Tue, Apr 06, 2004 at 05:57:54PM -0400, Gene Heskett wrote:
> >That's not quite my point. I am not afraid of running my athlons at
> >70C. I just don't want to. With Debian Woody they ran at <40C,
> > which is impressing IMHO. An upgrade to Sarge raised the temp for
> > about 5K, which is still very cool. This temperature didn't change
> > when I upgraded to an early 2.6 kernel. Just after 2.6.3-mm4 there
> > was this jump for 10K that I just do not understand. It doesn't
> > hurt the athlons but seems unnecessary to me.
> >
> >-jo
>
> 40C? Shut down for an hour to cool, I've never seen the post on my
> board show less than 63C by the time it gets to that part of the
> bios. I'm running a 1400DX at 1400mhz, so the bios thinks its a
> 1600DX, and I've got vcore set down to 1.65 volts which helps a bit.
Do you use anything besides a fan to keep your processors cool? I've no
experience with athlons on UP machines, but amd76x_pm does a good job on
MPs. I'm not joking: lmsensors sometimes reports 38C on low load.
>
> Actually, the athlons seem to have a builtin shutdown at 75C, I've hit
> that once or 3 times when the air under the desk was trapped worse
> than usual. Makes for downright ugly reboots...
Seems like athlon 2000+ MPs are more robust. Playing cube rises
temperature to 78C without any problems.
-jo
--
-rw-r--r-- 1 jo users 80 2004-04-07 18:02 /home/jo/.signature
In linux.kernel you write:
>At least on A7M266-D lmsensors read thermal sensors very wrong. I
>haven't got time to contact devs with that, but I do know for sure that
>amd76x_pm really does make cooling calls, even in 2.6.5-rc3-mm3
>(There should be /sys/devices/pci0000\:00/0000\:00\:00.0/C2_cnt file,
>which tells how many times has amd76x_pm really made the disconnection call).
lmsensors might be correct here as there is a sample config from Tyan
for my board. C2_cnt shows that amd76x_pm is indeed working.
Otherwise I had expected a much higher temperature.
>One issue is that from some kernel version amd76x_pm's idle() is called
>upto 3.5x times more often when there's some audio activity. So in
>effect number of calls to default_idle() jumps from 1100Hz to 3800Hz.
>(this is reproducible with 'rhytmbox' -application, but not with xmms.
>AFAIK my xmms uses OSS emulation and rhytmbox is native alsa.)
>Ahem. Could you actually try:
>echo 3 > /sys/devices/pci0000\:00/0000\:00\:00.0/lazy_idle
Did that.
>This could help gaining 5-8?C. HZ changed from 100 to 1000 in 2.6, so
>amd76x_pm old default doesn't apply overly well here.
Bingo! Currently running 2.6.5-mm1 at 42.0/43.5?C!
>There's some funniness going on with this tunable. It doesn't really
>affect how many times/second we call amd76x_pm.idle(), but rather how
>easily we go into sleep (no sleep if both CPU's aren't idle).
>With lazy_idle at 3 I get bad distortions with bttv card. with 3000 they
>disappear, but so does the thermal throttling :)
>(Sorry for lack of coherence right now)
Wouldn't be a bad idea to document this :-)
Thanks!
-jo
--
-rw-r--r-- 1 jo users 80 2004-04-07 19:37 /home/jo/.signature