|
Home > Archive > Red Hat Topics > February 2006 > Quad Opteron running at half the speed?
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
Quad Opteron running at half the speed?
|
|
|
| I just noticed that my Quad Opteron server with 2GHz CPU's is running at 1002MHz.
I could not believe my eyes when I saw /proc/cpuinfo returning
"cpu MHz : 1002.420"
The MoBo is TYAN, latest BIOS. Kernel 2.6.9-22.0.2.ELsmp but I can't see how kernel could
have anything to do with it. I would bet that originally the machine ran at 2 GHz.
Any ideas as to what could have happened? I cannot take the machine down to play with it
unless I have some sort of a plan. Mind boggles...
Thanks
<feeb@chem.utoronto.ca>
| |
| General Schvantzkoph 2006-02-01, 8:46 pm |
| On Wed, 01 Feb 2006 17:53:22 -0500, FEEB wrote:
> I just noticed that my Quad Opteron server with 2GHz CPU's is running at 1002MHz.
> I could not believe my eyes when I saw /proc/cpuinfo returning
> "cpu MHz : 1002.420"
>
> The MoBo is TYAN, latest BIOS. Kernel 2.6.9-22.0.2.ELsmp but I can't see how kernel could
> have anything to do with it. I would bet that originally the machine ran at 2 GHz.
>
> Any ideas as to what could have happened? I cannot take the machine down to play with it
> unless I have some sort of a plan. Mind boggles...
>
> Thanks
>
>
>
> <feeb@chem.utoronto.ca>
Do Opteron's support Cool and Quiet? The Athlon64s can run at reduced
clock rates when they aren't busy which reduces power and heat. Chances
are the Opteron can do the same thing. The clock speed is controlled by
the speed governors which are found in the cpufreq directories (cpux is
the cpu0, cpu1,cpu2,cpu3),
/sys/devices/system/cpu/cpux/cpufreq/scaling_available_governors
/sys/devices/system/cpu/cpux/cpufreq/scaling_available_frequencies
/sys/devices/system/cpu/cpux/cpufreq/scaling_governor
You need to upgrade to a modern kernel to get cpu frequency control. 2.6.9
lacked decent support for AMD64 power governors, 2.6.12.x and higher have
speed governor support. Upgrade to the current 2.6.15.2 kernel. If you
build it yourself make sure that you build in the speed governors,
# CPU Frequency scaling
#
CONFIG_CPU_FREQ=y
CONFIG_CPU_FREQ_TABLE=y
CONFIG_CPU_FREQ_DEBUG=y
CONFIG_CPU_FREQ_STAT=m
CONFIG_CPU_FREQ_STAT_DETAILS=y
# CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not set
CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE=y
CONFIG_CPU_FREQ_GOV_PERFORMANCE=y
CONFIG_CPU_FREQ_GOV_POWERSAVE=y
CONFIG_CPU_FREQ_GOV_USERSPACE=y
CONFIG_CPU_FREQ_GOV_ONDEMAND=y
CONFIG_CPU_FREQ_GOV_CONSERVATIVE=y
The 32 bit versions of Fedora don't include the speed governors in the
standard install (there might be an RPM, I don't know because I always
build my own kernels). The 64 bit version of FC4 does include the speed
governors.
| |
| Scott Lurndal 2006-02-01, 8:46 pm |
| "FEEB" <feeb@chem.utoronto.ca> writes:
>I just noticed that my Quad Opteron server with 2GHz CPU's is running at 1002MHz.
>I could not believe my eyes when I saw /proc/cpuinfo returning
>"cpu MHz : 1002.420"
>
>The MoBo is TYAN, latest BIOS. Kernel 2.6.9-22.0.2.ELsmp but I can't see how kernel could
>have anything to do with it. I would bet that originally the machine ran at 2 GHz.
>
>Any ideas as to what could have happened? I cannot take the machine down to play with it
>unless I have some sort of a plan. Mind boggles...
The opterons have both frequency and voltage scaling. On redhat distributions
the daemon cpuspeed will utilize frequency and voltage scaling to reduce power
usage and heat generation during idle cpu periods.
root# chkconfig cpuspeed off
root# service cpuspeed stop
will disable it.
Note that as the load on the system increases, cpuspeed will automatically
bump FID and VID up to the maximum values, but there will be some lag.
scott
| |
|
| On Thu, 02 Feb 2006 01:03:38 GMT, Scott Lurndal wrote:
>"FEEB" <feeb@chem.utoronto.ca> writes:
could[vbcol=seagreen]
it[vbcol=seagreen]
>
>The opterons have both frequency and voltage scaling. On redhat distributions
>the daemon cpuspeed will utilize frequency and voltage scaling to reduce power
>usage and heat generation during idle cpu periods.
>
>root# chkconfig cpuspeed off
>root# service cpuspeed stop
>
>will disable it.
>
>Note that as the load on the system increases, cpuspeed will automatically
>bump FID and VID up to the maximum values, but there will be some lag.
>
>scott
Thanks for the pointer. Unfortunately, RHEL distribution does not have man page for
cpuspeed and I could not find anything on the Web. Could you please point me to some
reading about cpuspeed and /etc/cpuspeed.conf?
Thanks
<feeb@chem.utoronto.ca>
| |
| Jan Gerrit Kootstra 2006-02-02, 5:47 pm |
| FEEB wrote:
> On Thu, 02 Feb 2006 01:03:38 GMT, Scott Lurndal wrote:
>
>
>
> could
>
>
> it
>
>
>
> Thanks for the pointer. Unfortunately, RHEL distribution does not have man page for
> cpuspeed and I could not find anything on the Web. Could you please point me to some
> reading about cpuspeed and /etc/cpuspeed.conf?
>
> Thanks
>
>
>
> <feeb@chem.utoronto.ca>
>
>
>
>
cpuspeed -?
| |
| General Schvantzkoph 2006-02-02, 5:48 pm |
| On Wed, 01 Feb 2006 23:24:42 -0500, FEEB wrote:
> On Thu, 02 Feb 2006 01:03:38 GMT, Scott Lurndal wrote:
>
> could
> it
>
> Thanks for the pointer. Unfortunately, RHEL distribution does not have man page for
> cpuspeed and I could not find anything on the Web. Could you please point me to some
> reading about cpuspeed and /etc/cpuspeed.conf?
>
> Thanks
>
>
>
> <feeb@chem.utoronto.ca>
Which version of RHEL are you using? The 2.6.9 kernel is very old and it
doesn't handle CPU speed scaling. You need to update your kernel to at
least a 2.6.12.x kernel although I'd recommend that you update to a 2.6.15
kernel if you can.
| |
| Jean-David Beyer 2006-02-02, 5:48 pm |
| General Schvantzkoph wrote:
> On Wed, 01 Feb 2006 23:24:42 -0500, FEEB wrote:
>
>
> Which version of RHEL are you using? The 2.6.9 kernel is very old and it
> doesn't handle CPU speed scaling. You need to update your kernel to at
> least a 2.6.12.x kernel although I'd recommend that you update to a 2.6.15
> kernel if you can.
>
From the kernel number, he is running the latest RHEL 4, which is Red Hat's
Latest.
the -22.0.2 is Red Hat's suffix: they put the patches in without changing
the 2.6.x part because their stability policy is to never put in new
features, but only to do bug and security fixes.
--
.~. Jean-David Beyer Registered Linux User 85642.
/V\ PGP-Key: 9A2FC99A Registered Machine 241939.
/( )\ Shrewsbury, New Jersey http://counter.li.org
^^-^^ 10:20:00 up 13 days, 1:47, 5 users, load average: 5.47, 4.79, 4.37
| |
| Steve Thompson 2006-02-02, 5:48 pm |
|
On Thu, 2 Feb 2006, Scott Lurndal wrote:
> The opterons have both frequency and voltage scaling. On redhat distributions
> the daemon cpuspeed will utilize frequency and voltage scaling to reduce power
> usage and heat generation during idle cpu periods.
>
> root# chkconfig cpuspeed off
> root# service cpuspeed stop
>
> will disable it.
>
> Note that as the load on the system increases, cpuspeed will automatically
> bump FID and VID up to the maximum values, but there will be some lag.
We have had many issues with cpuspeed. I have found, and this is exactly
the same on the many systems (10+) on which we have tried it, that simply
stopping the service does not restore the "proper" clock rate; it simply
leaves it at whatever it was, and furthermore, this lower setting persists
across reboots (unless the box is power cycled). Sending a USR1 signal to
cpuspeed prior to stopping it restores the clock rate. I have also never
seen the clock rate increase under load once it has been reduced.
Steve
| |
|
| On Wed, 01 Feb 2006 17:53:22 -0500, FEEB wrote:
> I just noticed that my Quad Opteron server with 2GHz CPU's is running at 1002MHz.
> I could not believe my eyes when I saw /proc/cpuinfo returning
> "cpu MHz : 1002.420"
>
> The MoBo is TYAN, latest BIOS. Kernel 2.6.9-22.0.2.ELsmp but I can't see how kernel could
> have anything to do with it. I would bet that originally the machine ran at 2 GHz.
>
> Any ideas as to what could have happened? I cannot take the machine down to play with it
> unless I have some sort of a plan. Mind boggles...
>
> Thanks
>
>
>
> <feeb@chem.utoronto.ca>
Various boards have the capability to set system speed - have you checked
this possibility in the BIOS setup?
| |
|
| On Thu, 02 Feb 2006 09:35:45 -0500, General Schvantzkoph wrote:
>On Wed, 01 Feb 2006 23:24:42 -0500, FEEB wrote:
>
at 1002MHz.[vbcol=seagreen]
see how kernel[vbcol=seagreen]
ran at 2 GHz.[vbcol=seagreen]
down to play with[vbcol=seagreen]
distributions[vbcol=seagreen]
power[vbcol=seagreen]
automatically[vbcol=seagreen]
man page for[vbcol=seagreen]
point me to some[vbcol=seagreen]
>
>Which version of RHEL are you using? The 2.6.9 kernel is very old and it
>doesn't handle CPU speed scaling. You need to update your kernel to at
>least a 2.6.12.x kernel although I'd recommend that you update to a 2.6.15
>kernel if you can.
>
I am using RHEL4U2 and I can assure you that cpuspeed works as it is
supposed to.
<feeb@chem.utoronto.ca>
| |
|
| Reply-To: "FEEB" <feeb@chem.utoronto.ca>
NNTP-Posting-Host: frank.chem.utoronto.ca
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
X-Trace: news1.chem.utoronto.ca 1138904317 13267 142.150.224.105 (2 Feb 2006 18:18:37 GMT)
X-Complaints-To: news@chem.utoronto.ca
NNTP-Posting-Date: Thu, 2 Feb 2006 18:18:37 +0000 (UTC)
X-Newsreader: PMINews 2.00.1205 For OS/2
Xref: number1.nntp.dca.giganews.com comp.os.linux.hardware:324760 comp.os.linux.redhat:44018 linux.redhat.misc:132853
On Thu, 02 Feb 2006 15:23:41 GMT, Jean-David Beyer wrote:
>General Schvantzkoph wrote:
at 1002MHz.[vbcol=seagreen]
see how kernel[vbcol=seagreen]
ran at 2 GHz.[vbcol=seagreen]
down to play with[vbcol=seagreen]
distributions[vbcol=seagreen]
reduce power[vbcol=seagreen]
automatically[vbcol=seagreen]
man page for[vbcol=seagreen]
point me to some[vbcol=seagreen]
2.6.15[vbcol=seagreen]
>From the kernel number, he is running the latest RHEL 4, which is Red
Hat's
>Latest.
>
>the -22.0.2 is Red Hat's suffix: they put the patches in without changing
>the 2.6.x part because their stability policy is to never put in new
>features, but only to do bug and security fixes.
Exactly. cpuspeed works just fine. I would just prefer to be able to read
something about it.
<feeb@chem.utoronto.ca>
| |
|
| On Thu, 02 Feb 2006 16:02:25 GMT, Steve Thompson wrote:
>
>On Thu, 2 Feb 2006, Scott Lurndal wrote:
>
distributions[vbcol=seagreen]
power[vbcol=seagreen]
automatically[vbcol=seagreen]
>
>We have had many issues with cpuspeed. I have found, and this is exactly
>the same on the many systems (10+) on which we have tried it, that simply
>stopping the service does not restore the "proper" clock rate; it simply
>leaves it at whatever it was, and furthermore, this lower setting persists
>across reboots (unless the box is power cycled). Sending a USR1 signal to
>cpuspeed prior to stopping it restores the clock rate. I have also never
>seen the clock rate increase under load once it has been reduced.
>
>Steve
I did some testing (I can't take the machine down, it's my production
server) and it works quite well here. It is capable of adjusting clock
rate per single CPU, so I've seen various combinations of clock rates on my
4 CPU's. When I start something CPU intensive (like dnetc), the clock rate
goes to max on all CPU's almost instantaneously. When dnetc is killed it
goes down right away. It reacts well to changing load on the machine
adjusting clock rate on any number of CPU's if required.
For your reference, this is my /etc/cpuspeed.conf:
VMAJOR=1
VMINOR=1
OPTS="$OPTS -n"
Unfortunately I have no idea what it does as I was not able to find any
documentation so far and I did not have time to poke through the source.
<feeb@chem.utoronto.ca>
| |
|
| On Thu, 02 Feb 2006 15:16:45 +0100, Jan Gerrit Kootstra wrote:
>FEEB wrote:
at 1002MHz.[vbcol=seagreen]
see how kernel[vbcol=seagreen]
ran at 2 GHz.[vbcol=seagreen]
down to play with[vbcol=seagreen]
distributions[vbcol=seagreen]
power[vbcol=seagreen]
automatically[vbcol=seagreen]
man page for[vbcol=seagreen]
point me to some[vbcol=seagreen]
>cpuspeed -?
That's the daemon that does clock rate and voltage adjustments in RHEL4U2.
It works quite impressively.
<feeb@chem.utoronto.ca>
| |
| Jan Gerrit Kootstra 2006-02-02, 5:48 pm |
| FEEB wrote:
> On Thu, 02 Feb 2006 15:16:45 +0100, Jan Gerrit Kootstra wrote:
>
>
>
> at 1002MHz.
>
>
> see how kernel
>
>
> ran at 2 GHz.
>
>
> down to play with
>
>
> distributions
>
>
> power
>
>
> automatically
>
>
> man page for
>
>
> point me to some
>
>
>
> That's the daemon that does clock rate and voltage adjustments in RHEL4U2.
>
> It works quite impressively.
>
>
> <feeb@chem.utoronto.ca>
>
>
>
>
FEEB,
cpuspeed -? is a way to find out more about the options of the command.
Kind regards,
Jan Gerrit Kootstra RHCE
| |
|
| On Thu, 02 Feb 2006 19:55:19 +0100, Jan Gerrit Kootstra wrote:
>FEEB,
>
>
>cpuspeed -? is a way to find out more about the options of the command.
Actually, here it is 'cpuspeed -h'. The above does not work here.
Cheers
<feeb@chem.utoronto.ca>
|
|
|
|
|