Idle power usage on bare metal higher with fresh Proxmox VE 8 install than Windows 11

J

jolystep

New Member
Jan 11, 2025
5
0
1
I installed proxmox on an Asrock Z790M-ITX wifi motherboard (mini-itx) with a i5-13500, 32gb of ram, 2 nvme SSDs, and a 3090.

Idle power is running a fresh proxmox install with no guests at all at approximately 100W. The exact same hardware gets me 60W running windows 11 (with a running ollama server). Is there something I am missing as to why a fresh proxmox install runs with so much more idle power than the windows 11 install? no difference in bios settings.

I would really like for proxmox to get under 60W if possible while also running ollama but if thats not possible than keeping this machine as a windows machine may be the only option for me and my homelab. Low power draw is a priority for me.

Would love some advice/or thoughts if anyone has similar experience. I couldn't find any threads on this pertaining to bare-metal installs. Most of the idle power questions were about when guests were running on the proxmox install.
 
ProxmoxVE by default always runs with "highest performance". Windows by default is using something like balanced power profile.
So this is expectet. ProxmoxVE is a datacenter server virtualisation. A location where expect maximum performance at all time.

You might check if tunedadm with a balanced power profile might fit your home needs.
 
  • Like
Reactions: jolystep
That makes complete sense. Thanks @itNGO for the quick reply. I will look into tunedadm.

Is that the only way to manage the power profile on a pve node? It would be really awesome if proxmox could build that kind of configuration into the GUI for enthusiasts/hobbyists like me. I'm definitely not afraid of the command line so I'm willing to explore more complicated procedures.
 
It depend on your real Mainboard, CPU and the power save and control settings.
So no GUI configuration s available or necessary.
Proxmox VE is not a enthusiasts/hobbyists "tool" in my opinion.
The main ore most settings work only in the CLI background, it's Linux like ok for me.
 
So i tried tuned-adm and got a grand total of 4W saved on the "powersave" profile. Still quite far from Windows 11's 60W at idle.

Unless anyone else has had success with bringing the idle power (with no guests at all) down more significantly than that, I think I might have my answer :(
 
powertop is showing me something interesting

Code: 
PowerTOP 2.14     Overview   Idle stats   Frequency stats   Device stats   Tunables   WakeUp               


           Pkg(OS)  |            Core(OS) |            CPU(OS) 0   CPU(OS) 1
POLL        0.0%    | POLL        0.0%    | POLL        0.0%    0.0 ms  0.0%    0.0 ms
C1_ACPI     0.0%    | C1_ACPI     0.0%    | C1_ACPI     0.0%    0.3 ms  0.0%    0.7 ms
C2_ACPI    98.6%    | C2_ACPI    99.8%    | C2_ACPI    99.7%   45.9 ms 99.9%   60.9 ms
C3_ACPI     0.0%    | C3_ACPI     0.0%    | C3_ACPI     0.0%    0.0 ms  0.0%    0.0 ms

                    |            Core(OS) |            CPU(OS) 2   CPU(OS) 3
                    | POLL        0.0%    | POLL        0.0%    0.0 ms  0.0%    0.0 ms
                    | C1_ACPI     0.0%    | C1_ACPI     0.1%    0.2 ms  0.0%    0.0 ms
                    | C2_ACPI    97.5%    | C2_ACPI    99.2%   31.9 ms 95.7%  710.5 ms
                    | C3_ACPI     0.0%    | C3_ACPI     0.0%    0.0 ms  0.0%    0.0 ms

                    |            Core(OS) |            CPU(OS) 4   CPU(OS) 5
                    | POLL        0.0%    | POLL        0.0%    0.0 ms  0.0%    0.0 ms
                    | C1_ACPI     0.0%    | C1_ACPI     0.1%    0.5 ms  0.0%    0.0 ms
                    | C2_ACPI    96.8%    | C2_ACPI    97.9%   47.2 ms 95.7%  872.0 ms
                    | C3_ACPI     0.0%    | C3_ACPI     0.0%    0.0 ms  0.0%    0.0 ms

                    |            Core(OS) |            CPU(OS) 6   CPU(OS) 7
                    | POLL        0.0%    | POLL        0.0%    0.0 ms  0.0%    0.0 ms
                    | C1_ACPI     0.0%    | C1_ACPI     0.0%    0.2 ms  0.0%    0.2 ms
                    | C2_ACPI    97.3%    | C2_ACPI    98.9%   58.0 ms 95.7%  145.3 ms

<ESC> Exit | <TAB> / <Shift + TAB> Navigate |


seems like cstates are not enabled. i got this idea from Claude who suggested enabling power saving in GRUB + forcing ASPM with these commands

# Edit /etc/default/grub and add these parameters to GRUB_CMDLINE_LINUX_DEFAULT:
GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_pstate=enable processor.max_cstate=8 intel_idle.max_cstate=8"

# Update GRUB
update-grub

echo "options pcie_aspm=force" > /etc/modprobe.d/pcie_aspm.conf

will give this a go and report back
 
You must log in or register to reply here.
Top Bottom