Two seemingly identical Windows 10 VMs. 20% vs 2% host CPU while VM idle

[marcosscriven]

Host: Proxmox 8.0.4, Ryzen 5700x, 64GB
Both Guest VMs: Windows 10, all latest updates. 32 GB assigned. Both with 'host' CPU type. Both same virtual SSD settings

When I run one of the VMs, and let it settle down with task manage open, I see it idling as expected. Looking at 'top' on the host I see max CPU usage of 2-4%.


If I stop that VM and start the other, it shows 20-25% CPU usage for one KVM thread.

I suppose it could be something to do with other software in the VM, however if both guest task managers show idle, I'm not sure what else would be using CPU.

Here's the config of the two:

Higher idle power VM:

agent: 1
balloon: 0
bios: ovmf
boot: order=scsi0;net0
cores: 16
cpu: host
description: scsi1%3A local-fast%3Avm-110-disk-2,backup=0,cache=writeback,discard=on,iothread=1,size=512G,ssd=1
efidisk0: local-fast:vm-110-disk-0,efitype=4m,pre-enrolled-keys=1,size=4M
hostpci0: 0000:08:00,pcie=1,x-vga=1
hostpci1: 0000:0a:00.3,pcie=1,rombar=0
machine: pc-q35-8.0
memory: 32768
meta: creation-qemu=8.0.2,ctime=1690043430
name: windows10-gaming
net0: virtio=,bridge=vmbr0,firewall=1
numa: 0
ostype: win10
scsi0: local-fast:vm-110-disk-1,cache=writeback,discard=on,iothread=1,size=64G,ssd=1
scsi1: local-fast:vm-110-disk-2,backup=0,cache=writeback,discard=on,iothread=1,size=512G,ssd=1
scsihw: virtio-scsi-single
smbios1: uuid=...
sockets: 1
tablet: 0
vmgenid: ...

This results in the CPU using 15-20W more, presumably as it never gets into a low power state.


Low idle power:

qm config 111
agent: 1
balloon: 0
bios: ovmf
boot: order=scsi0;net0
cores: 16
cpu: host
efidisk0: local-lvm:vm-111-disk-0,efitype=4m,pre-enrolled-keys=1,size=4M
hostpci0: 0000:0a:00.3
hostpci1: 0000:08:00,pcie=1,rombar=0,x-vga=1
machine: pc-q35-8.0
memory: 32768
meta: creation-qemu=8.0.2,ctime=1690043430
net0: virtio=,bridge=vmbr0,firewall=1
numa: 0
ostype: win10
scsi0: local-lvm:vm-111-disk-1,cache=writeback,discard=on,iothread=1,size=64G,ssd=1
scsihw: virtio-scsi-single
smbios1: ...
sockets: 1
vmgenid: ...

 

[Anotheruser]

Some kind of background task?
Did you check the details page of the task manager, this gives you a more detailed list of whats running and using resources?

 

[leesteken]

Why the differences in the configuration? Why rombar=0 for 08:00 in one and for 0a:00.3 in the other? Why tablet:0 (use tablet for pointer) in one but not the other? Especially that last one can cause some CPU usage according to some threads here.

 

[marcosscriven]

Some kind of background task?
Did you check the details page of the task manager, this gives you a more detailed list of whats running and using resources?

Thanks - but there's nothing running with more than 0 to 1% total, and I checked the details page.

 

[marcosscriven]

Why the differences in the configuration? Why rombar=0 for 08:00 in one and for 0a:00.3 in the other? Why tablet:0 (use tablet for pointer) in one but not the other? Especially that last one can cause some CPU usage according to some threads here.

Thanks for response.

1. The rombar differences were kind of random - neither actually needs it (08:00 is a UEFI GPU, 0a:00.3 is a UBS controller). In case somehow it did make a difference, I just tried, and I get the same results

2. Yes I specifically tried turning off the tablet pointer on the one with higher idle resource usage having seen, as you pointed out, that it can sometimes cause CPU usage with polling. It didn't help however.

So I've still no idea why, which is frustrating.

 

[leesteken]

So I've still no idea why, which is frustrating.

Can you boot both VMs with an Ubuntu Live Installer ISO? Just start it but don't install Ubuntu. See if you get a different power usage on an idle graphical desktop. That could rule out Proxmox and you would have to focus on what Windows is doing differently.

 

[marcosscriven]

Can you boot both VMs with an Ubuntu Live Installer ISO? Just start it but don't install Ubuntu. See if you get a different power usage on an idle graphical desktop. That could rule out Proxmox and you would have to focus on what Windows is doing differently.

Interesting idea - I'll try that and report back. Thanks!