XPost: alt.comp.os.windows-11, alt.msdos.batch
On Tue, 9/2/2025 3:55 PM, John B. Smith wrote:
I hope I'm not highjacking the thread if I start asking about wattage measurements? I'll start another thread if that's needed. I'd never
checked wattage on my i6 chip Asus Z790 motherboard before so this
thread got me started.
I have 53 background processes running in my Win11 Taskmanager. I
don't know how to safely turn off the ones running to make
measurements. Just idling along the KWMeter is reading watts from 53.3
to 136. CoreTemp reading watts from 7.9 to 19.8
On the boot I see 119 to 214 watts.
Machine shut down, reads 1.3 to 1.4 watts.
Can you tell me a safe way to stop those processes in Taskmaster that
aren't needed without crashing the machine?
The SVCHOST, most of them use *zero* cycles. They're not
scheduled to run and they don't run. They are "on-demand"
services.
Running Sysinternals Process Explorer as administrator, there is
an interface that gives a cycle count per process. That's how you
can check at that level.
A few are hopefuls. WUAUSERV for Windows Update, it would run
several times a day, and it can run for a bit, preparing
WinSxS for maintenance. That's a fairly expensive one, in
terms of KwH.
The SysMain has likely been renamed a number of times. It
positions materials on a HDD for faster loading (.pf prefetch files).
On an SSD, this is likely a waste of a service. During a Windows
update, I run services.msc and I shut that one down. I
also try to kill SearchIndexer (a separate process, not
a SVCHOST), but that one is the undead and it keeps starting
itself.
But for the most part, the consumption in the OS is going to
consist of other maintenance things. These would be forked,
run for a while, and close on their own. These like to start
when you're trying to benchmark and stuff.
When you ask for a compute job to be done, the job has
"a fixed number of watt-hours" in a sense. Stretching the
job out, doesn't make a lot of sense, so for the
following, it's better to "rush the job, then turn
the PC OFF". But if the machine must run all day, you
can try and crank it down a bit. It depends on how
"overpowered" the resources are, how much room there is
to do this.
Some motherboards have an "eco mode". My 5950X, that mode
is simple -- turn one of the two silicon die off. My other
processors have a single-die, and an "eco mode" consists of
running reduced frequency (and reduced VCore voltage) on the processor.
CPUs have closed loop controls, like the video cards got. If
you set the BIOS CPU power limiter to "100W", then it starts to
throttle when the power gets that high, and that is an
automatic way of capping top power. The Intel one goes to 4095
watts, which means "unlimited" in a sense. That might help
if your cooler isn't very good.
You can turn the core count down in the BIOS, on some boards.
Some CPUs may not have such controls. A Phenom II from AMD,
the six die were in two groups, and you could shut one group
off (even though they shared a common die).
One very expensive processor ($1000), if you shut it down to 1 core
in the BIOS, right after you save settings... the CPU is
destroyed. That's an example of a CPU where you *never* drop
to 1 core (setting not blocked in BIOS!). 2 cores would be OK
(of six perhaps). These probably aren't power modulated, just
a clock thing (reducing clock to a processor, reduces consumption
to DC leakage). When the BIOS detects one of your attempts to
save power, it still tries to arrange things so the die is as
"power balanced" as possible. The ECL designer in my group at
work used to do that for the chips he worked on, power-balance
the corners so the die heats evenly and doesn't crack at
extremely high temps (ECL loves heat).
But messing with things at that level, ECO modes, power capping,
that makes more of an impact than doing a lot of tweaking
at individual process levels, especially when a lot of the
SVCHOST are drawing zero cycles anyway.
Microsoft has done a lot already, to optimize things. They have
had studies running on machines, to find things they can turn down.
On some motherboards, it is the *chipset* which is a pig. The
CPU gets blamed, but the Idle CPU actually draws less than a
Northbridge with 42 PCIe lanes on it. While you can turn
chipset voltage down, on those pig chips, fiddle too much
and you get memory errors, so it's not really worth
chasing that. As it is, the pig chip across the way has
a boost of a tenth of a volt to keep it "honest". My X48,
that was a badly binned chip, and that needed a boost its
whole life and it probably should never have been put in
the "good" bin at Intel. That chipset is dead now. The X79
is still going strong (and drinking the power like crazy).
Buying a modern computer can save power, but the payback
period would be longer than you will be alive. Like using
my 33W computer, instead of the 100W idle computer, that
saves money. Pennies a day. It's because of this, the 33W
computer tends to run all day, the higher end machines
stay OFF. But if I wanted to compress a 3TB Macrium MRIMG,
I wouldn't do that on this machine (would take too long).
On the 5950X, railing one core uses half the CPU power
and ramping up all the cores and threads, doubles that...
200W ____
___/
___/
100W ___/
/
/
33W /
0 1 N
That is why saving power is so hard, as vampire tasks
that use a fraction of a core, they tend to climb the
steep part of the curve near the "one core railed" end.
An ECO setting with an artificially capped Fmax, will
likely scale the whole graph. At least some of the load
at the low end, is ATX supply inefficiency, chipset
static power (power that does not change with usage),
you charging your iPhone and so on.
100W ____
___/
___/
50W ___/
/
/
33W /
0 1 N
Paul
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)
