CPU scaling - why aren't I drawing more power?

[tppytel]

Longtime Debian user (since ~2000), but I haven't done a fresh install in several years at least. (When did we get real web forums instead of mailing lists? How chic!) I'm configuring a new home server and exploring current power saving technology. Hardware is a new Kaby Lake i3 (TDP 51W) on an Asus H170 board, running Jessie with the backported 4.x kernel. The box is hooked up to a Kill-A-Watt on my outlet which reports power usage of ~21W after installation, which seems fine. But I was playing around with the various cpufreq and kernel options just to learn how the architecture works, and I don't understand what I'm seeing. I was trying to run the CPU cores at full speed just to see what the max power draw would be, but the reading at the Kill-A-Watt doesn't change from ~21W. I booted with intel_pstate=disable and ran "cpufreq-set -g performance". Also disabled Asus's EPU in the BIOS. Running cpufreq-info reports the current core speeds are at the max of 3.9GHz on all cores. Cat'ing the various files in /sys shows the performance governor in use and the max frequency being used as well. So why doesn't the reading at the Kill-A-Watt change? I was expecting it to increase significantly.

Obviously this isn't a "real" problem - I don't really want to use more power than I need to. But I do want to understand how the current power management system actually works. Is something else scaling CPU frequencies down or reducing power draw here?

[phenest]

I have one of those outlet devices for measuring power consumption too. But 21W doesn't seem much to me. Are you sure it's working correctly? I'd imagine a low end consumption of at least 100W.

[tppytel]

This is the first time I've used the Kill-A-Watt myself, so I don't really know. But there's not much in the system - a pair of solid state drives, integrated graphics that are just driving a text console, and a couple of low-speed fans. There's also a mechanical drive and an additional NIC, but neither of those are being used by the OS yet. If I load the CPU using the "stress" tool, the KAW registers ~50W consumption.

[Head_on_a_Stick]

I booted with intel_pstate=disable and ran "cpufreq-set -g performance".

Have you tried the "performance" setting with pstate enabled?

It should still work: https://www.kernel.org/doc/Documentatio ... pstate.txt

Also, I would submit that perhaps the information in syphilisfs, sorry, sysfs is incorrect.

The idea that frequency can be set to a single frequency is fictional for Intel Core processors. Even if the scaling driver selects a single P-State, the actual frequency the processor will run at is selected by the processor itself.

[tppytel]

Setting the performance governor with P-states enabled does not increase measured power consumption either.

The idea that frequency can be set to a single frequency is fictional for Intel Core processors. Even if the scaling driver selects a single P-State, the actual frequency the processor will run at is selected by the processor itself.

But that would explain things. The CPU is smart enough not to run any faster than it actually needs to. Thanks.

[pendrachken]

Did you measure the draw when the system was actually under load? The totals for watt draw are usually for when the CPU is under load and working, not sitting idle... no matter the clock scaling.