Intel_Idle C-States in Proxmox with Intel Series 200 CPUs

bastae

New Member
Jan 12, 2026
2
0
1
Hi everyone,

I’m looking for experiences or insights regarding CPU C-States on Intel Core 200 Series CPUs, specifically the Intel Core Ultra 245K.

My setup:

  • CPU: Intel Core Ultra 245K
  • Mainboard: Gigabyte B860M DS3H
  • OS: Linux (PROXMOX)
  • Additional hardware: 10G NIC + 3 HDDs (ZFS) + 1 SSD NVME
The issue I’m facing is that despite the intel_idle driver being active, Powertop only reports C1 / C3 ACPI states. I don’t see any deeper C-States from the intel_idle driver, in Powertop.

Additionally, my idle power consumption is around 55 W, which seems quite high to me, even considering:

  • A 10G NIC
  • 3 HDDs in a ZFS pool
All devices report ASPM support, and ASPM is enabled both in BIOS and in the OS.

So my questions are:

  • Has anyone successfully achieved deeper C-States on Intel Core Ultra 200 Series CPUs?
  • Is this a known limitation with the B860 chipset or specifically with Gigabyte boards?
  • Are there known BIOS settings, kernel parameters, or firmware updates that help unlock deeper C-States on this platform?
I am close to loosing my mind over this. Any hints are appreciated!
 
More details, or otherwise there is no sense in continuing this topic:
  • What exact kernel version is currently used?
  • What exact 10G card do you have?
  • And finally what exact HDDs are you mentioning?

Losing your mind over a few extra Watts is not really worth it.

Besides the points above, did you already try another Linux distribution and had a look at the powertop output there?
Best would be something with the newest possible kernel. With mainline we are nearing the release of kernel 6.19, while the newest available kernel with proxmox currently seems to be 6.17. And the hardware is so new, that this very well might still make a difference.
 
  • Like
Reactions: bastae
Hey @celemine1gig, thank you for your answer!

Here are the requested details:
  • Kernel: 6.17.4-2-pve
  • HDDs: Western Digital 10TB Red Plus (in ZFS RAIDZ1)
  • NIC: TP-LINK TX410 10G PCIe (Aquantia AQtion AQC113, NBase-T / IEEE 802.3an)

To clarify: this is not about stressing over 5–10 watts.
My goal is to better understand system optimization in general, especially CPU C-states, the behavior of intel_idle, and how this relates to what Powertop reports. But if it helps with power consumption i wouldnt say no either...

I agree that the hardware is still quite new and that kernel versions may have a noticeable impact. I plan to test another distribution with a newer mainline kernel, although I’ll probably only get to that over the weekend. Do you have a favorite distro for trying these things?

Is there anything specific I should be careful about when doing this on a "production Proxmox system", if the goal is analysis only?
 
For the question of distribution for testing:
Due to the huge number of users, the availablity of a mainline linux kernel repository and the fact, that Proxmox actually uses a adjusted Ubuntu kernel, Ubuntu is what I would choose.
Just use the latest LTS-release (https://releases.ubuntu.com/noble/) and add the mainline kernel PPA (https://code.launchpad.net/~cappelikan/+archive/ubuntu/ppa). With that I would start experimenting.

Concerning the C-states, this is a very tricky topic, as lots of factors influence this.

Very simply put it is like this:
  • C-States are low power states for both the CPU-cores alone and also for the whole CPU-package (incluing all the parts besides the sole CPU cores, like memory controller, GPU, ...)
  • Each next lower C-State has some requirement, so that it can be reached/activated.
  • A C-state can result in a HW-part to be put into a deep sleep state, where it just is inactive (yet still powered), or it can even mean that the power-supply to that very HW-part is completely cut off.
  • As a result, each state has DIFFERENT prerequisites and also different wake-up times, in order to get back to a fully operational state, after entering that state.
Now, for example, if you have a SATA or NVMe disk, that for some reason does not support the required power management states (which Enterprise drives often do, due to not seeing the need for it), then this can block the controller (SATA or PCIe) from reaching a certain low-power mode and consequently holding back the whole CPU-package from reaching lower states as well, as that low-power state of the said controller would be a requirement for the package's lower C-State.

As a result, one can only reach the lowest C-States, if ALL involved components support the required power states and if all states are setup correctly in BIOS as well. Disabling ASPM for an end-device or a PCIe port in BIOS for example can also rain on your parade, concerning C-States.
You have to meticulously investigate each component of your system and its setup, in order to make it all work together for the best result.

The best/easiest way to tackle this problem, in my opinion, is to do it step-by-step.
Start with a setup, that is as minimal as possible.
For example, just board, CPU and RAM and have a look at how this behaves.
If this already has issues reaching lower power states, you don't even have to add any other hardware, before you solved those existing "issues".

Powertop has lots of indicators for what is going on (and wrong).
https://greencompute.uk/References/Power_Measurement/powertop_users_guide_201412.pdf
The "device stats"-tab for example will show you devices, that just stay 100% active all the time. And those are most likely the ones blocking the system from reaching lower power states.

Only once this minimal setup is working as intended, you go ahead and add the next single part.
Test again and look for possible issues.
Repeat...

This can be VERY time consuming, very much depening on your choice of hardware and also LARGELY on the BIOS setup.
And often times you realize in the end that some specific piece of hardware is ruining the whole nice low-power situation.
So, before you start with this you should most likely think about how much effort you want to put into this. This can easily get out of hand, both concerning time and also financials. :) Sadly speaking from experience here.