8:45 a.m.
I've got two questions related to cpu scaling. On my laptop I've got a
Core 2 Duo. Can the two cores scale frequencies independently? Most of
the time that doesn't seem like the case but every once in a while, my
cpufreq applets show different speeds. I'm assuming that there is just a
bit of slop involved and the two cores are always the same speed.
On my desktop, I've got Xeon CPUs that support limited scaling; they run
at either 2.8 or 3.4GHz. On FC5, they used to scale on demand but under
FC6 they don't and the cpufreq applet has a manual setting for 2.8 or
3.4 with an automatic setting that is greyed out.
Thanks,
tjb
--
=======================================================================
| Thomas Baker email: tjb(a)unh.edu |
| Systems Programmer |
| Research Computing Center voice: (603) 862-4490 |
| University of New Hampshire fax: (603) 862-1761 |
| 332 Morse Hall |
| Durham, NH 03824 USA http://wintermute.sr.unh.edu/~tjb |
=======================================================================
11:01 a.m.
On Mon, Oct 02, 2006 at 09:45:27AM -0400, Thomas J. Baker wrote:
I've got two questions related to cpu scaling. On my laptop I've got a
Core 2 Duo. Can the two cores scale frequencies independently?
No. Think about it. You have a single package, with a single power source.
Most of
the time that doesn't seem like the case but every once in a while, my
cpufreq applets show different speeds. I'm assuming that there is just a
bit of slop involved and the two cores are always the same speed.
It's a bug, but it should be cosmetic only. It's on the TODO, but
there are a number of other issues in this code that need tackling
first before it gets clean enough to fix this properly.
FC6-update/FC7 stuff.
On my desktop, I've got Xeon CPUs that support limited scaling;
they run
at either 2.8 or 3.4GHz. On FC5, they used to scale on demand but under
FC6 they don't and the cpufreq applet has a manual setting for 2.8 or
3.4 with an automatic setting that is greyed out.
Which cpufreq driver were they using ?
File a kernel bug, and attach dmesg.
Dave
11:16 a.m.
fedora-test-list-bounces(a)redhat.com wrote on 10/02/2006 12:01:15 PM:
On Mon, Oct 02, 2006 at 09:45:27AM -0400, Thomas J. Baker wrote:
>
> I've got two questions related to cpu scaling. On my laptop I've got a
> Core 2 Duo. Can the two cores scale frequencies independently?
No. Think about it. You have a single package, with a single power
source.
Regardless... I just ran KSysGuard, and had it plot the CPU frequency
for both halfs of my Core 2 Duo, and then ran 'find /' in one window.
What you see is that the clock on one CPU went up, and the other did not.
So based on that, I'd say yes... they can scale independently.
This document is strictly confidential and intended only for use by the addressee unless
otherwise stated. If you are not the intended recipient, please notify the sender
immediately and delete it from your system.
11:32 a.m.
On Mon, Oct 02, 2006 at 12:16:37PM -0400, Fulko.Hew(a)sita.aero wrote:
fedora-test-list-bounces(a)redhat.com wrote on 10/02/2006 12:01:15 PM:
> On Mon, Oct 02, 2006 at 09:45:27AM -0400, Thomas J. Baker wrote:
> >
> > I've got two questions related to cpu scaling. On my laptop I've got
a
> > Core 2 Duo. Can the two cores scale frequencies independently?
>
> No. Think about it. You have a single package, with a single power
source.
Regardless... I just ran KSysGuard, and had it plot the CPU frequency
for both halfs of my Core 2 Duo, and then ran 'find /' in one window.
What you see is that the clock on one CPU went up, and the other did not.
So based on that, I'd say yes... they can scale independently.
You snipped the part of my reply which explained..
So I'll add it back..
> It's a bug
Dave
--
http://www.codemonkey.org.uk
2:44 p.m.
On Mon, 2006-10-02 at 12:01 -0400, Dave Jones wrote:
On Mon, Oct 02, 2006 at 09:45:27AM -0400, Thomas J. Baker wrote:
>
> I've got two questions related to cpu scaling. On my laptop I've got a
> Core 2 Duo. Can the two cores scale frequencies independently?
No. Think about it. You have a single package, with a single power source.
> Most of
> the time that doesn't seem like the case but every once in a while, my
> cpufreq applets show different speeds. I'm assuming that there is just a
> bit of slop involved and the two cores are always the same speed.
It's a bug, but it should be cosmetic only. It's on the TODO, but
there are a number of other issues in this code that need tackling
first before it gets clean enough to fix this properly.
FC6-update/FC7 stuff.
> On my desktop, I've got Xeon CPUs that support limited scaling; they run
> at either 2.8 or 3.4GHz. On FC5, they used to scale on demand but under
> FC6 they don't and the cpufreq applet has a manual setting for 2.8 or
> 3.4 with an automatic setting that is greyed out.
Which cpufreq driver were they using ?
File a kernel bug, and attach dmesg.
Dave
Just curious - it must also a bug that under FC5, with two hyperthreaded
Xeon CPUs, all four logical processors scale independently?
I looked at my FC5 setup and had specified "DRIVER=acpi-cpufreq"
manually. Doing this for FC6 made the scaling behave like FC5,
automatically scaling up for load.
Thanks,
tjb
--
=======================================================================
| Thomas Baker email: tjb(a)unh.edu |
| Systems Programmer |
| Research Computing Center voice: (603) 862-4490 |
| University of New Hampshire fax: (603) 862-1761 |
| 332 Morse Hall |
| Durham, NH 03824 USA http://wintermute.sr.unh.edu/~tjb |
=======================================================================
3:01 p.m.
On Mon, Oct 02, 2006 at 03:44:02PM -0400, Thomas J. Baker wrote:
Just curious - it must also a bug that under FC5, with two
hyperthreaded
Xeon CPUs, all four logical processors scale independently?
Yes. They should scale in pairs. That is, if you're using a real
scaling driver, and not p4-clockmod. With p4-clockmod each
core is independant, because it's not really scaling, it's inserting
waitstates in each clock tick per 'thread'.
Dave
--
http://www.codemonkey.org.uk
3:16 p.m.
>>>> "DJ" == Dave Jones <davej(a)redhat.com>
writes:
DJ> With p4-clockmod each core is independant, because it's not really
DJ> scaling, it's inserting waitstates in each clock tick per
DJ> 'thread'.
Does p4-clockmod actually save any power, reduce heat output or do
anything useful other than making the CPU act slower?
- J<
3:19 p.m.
On Mon, Oct 02, 2006 at 03:16:26PM -0500, Jason L Tibbitts III wrote:
>>>>> "DJ" == Dave Jones
<davej(a)redhat.com> writes:
DJ> With p4-clockmod each core is independant, because it's not really
DJ> scaling, it's inserting waitstates in each clock tick per
DJ> 'thread'.
Does p4-clockmod actually save any power, reduce heat output or do
anything useful other than making the CPU act slower?
Allegedly it reduces heat by 'stretching' out a long running
job over a longer period of time.
Dave
--
http://www.codemonkey.org.uk
1:43 p.m.
On Mon, 2006-10-02 at 16:19 -0400, Dave Jones wrote:
On Mon, Oct 02, 2006 at 03:16:26PM -0500, Jason L Tibbitts III
wrote:
> >>>>> "DJ" == Dave Jones <davej(a)redhat.com> writes:
>
> DJ> With p4-clockmod each core is independant, because it's not really
> DJ> scaling, it's inserting waitstates in each clock tick per
> DJ> 'thread'.
>
> Does p4-clockmod actually save any power, reduce heat output or do
> anything useful other than making the CPU act slower?
Allegedly it reduces heat by 'stretching' out a long running
job over a longer period of time.
It's important to note that the rate of heat production will be lower
this way, but the total heat produced (i.e. actually energy used) during
the lifetime of a given job will be slightly higher.
That is, it doesn't save any power at all, and it produces slightly more
heat in the long run. But if you've got adequate HVAC you'll *notice*
less of the heat when you're sitting in front of the machine.
I.e. it sucks.
--
Peter
8:07 a.m.
On 10/6/06, Peter Jones <pjones(a)redhat.com> wrote:
It's important to note that the rate of heat production will be lower
this way, but the total heat produced (i.e. actually energy used) during
the lifetime of a given job will be slightly higher.
That is, it doesn't save any power at all, and it produces slightly more
heat in the long run. But if you've got adequate HVAC you'll *notice*
less of the heat when you're sitting in front of the machine.
The resistance of the wire does increase with temperature, so there
may be some advantages to running at a lower temperature for longer.
I'm not sure if this effect would be enough to reduce overall power
usage, but it might be.
n0dalus.
6437
days inactive
6442
days old
9 comments
6 participants
participants (6)
-
Dave Jones -
Fulko.Hew＠sita.aero -
Jason L Tibbitts III -
n0dalus -
Peter Jones -
Thomas J. Baker