Disabling Proxmox VE "Use tablet for pointer" option lowers idle CPU usage in VM
-
I read something about the "Use tablet for pointer" option in Proxmox VE that can lower idle CPU usage, so I experimented a bit last night. The link there says it helps Windows, though the only VM that I saw with a noticeable difference was a pfSense VM...not CentOS, Debian, or Windows. So maybe something specific to FreeBSD? Here's the CPU graph dropping from ~3% to ~0.5% after unchecking that option:

Thought this might be useful to someone. Of course idle CPU % will depend on other things like CPU model. The wiki page entry above links to an old forum thread.
-
What type of console is your VM using?
That setting can help in some cases, but it shouldn't be necessary if you use the SPICE console as the docs suggest:
https://docs.netgate.com/pfsense/en/latest/recipes/virtualize-proxmox-ve.html
-
@jimp Hmm, I read that doc note as "while using the console" which of course is rarely done on pfSense after install.
Datacenter > Options > Console Viewer is actually "Default (xterm.js)" but that's disabled on all VMs AFAIK and as a result I'm using noVNC on all VMs. Plus as I understood it SPICE required extra software and I didn't want to bother with that either.
Enabling SPICE on the VM requires a shutdown so I can try that some evening when it's not in use maybe.
-
It's not just while using the console, it's all the time. When I wrote the doc it had better results with the SPICE console method than the tablet pointer option which is why it wasn't mentioned on there. All of the options on that doc are the ones we recommend.
You can still use VNC with that console type if you want, the extra software is optional (but it's actually quite nice and worth using).
-
@jimp I tried on a Debian 13 and FWIW that had the opposite result. The graph image is skewed because there was actual CPU usage for the boots, but CPU was a pretty steady ~0.21%, with SPICE it settled around 0.4-0.6% for the next 10m or so, and after reverting it's around 0.3%. Maybe not the most scientific test, and obviously that's a different OS.
The "tablet" option doesn't require a boot, I believe. I did anyway for another reason but I believe my image above drops before that. "why not both?" I guess I'll see.
Thanks.
-
@jimp I tried SPICE tonight. Set display=SPICE, enabled "use tablet," shut down, start, let it sit several minutes. Did not actually view the console at any point. CPU at idle went from about 0.5% to 3.5-4%. After several more minutes (~7m uptime) I unchecked "use tablet" and it dropped to 0.5% again. I guess people can try it. Like I said it may depend on some other factor too. This VM is set at x86-64-v3, 3 cores. The arrow is about where I unchecked "use tablet" and the big spike is the boot (the graph takes a bit to update of course so it tapers down, and is 1m per line):


Also I did get to test SPICE on a Windows VM, no change that I could notice.
Edit: PVE 9.1.6
-
@SteveITS So I was curious, got a FreeBSD VM running on Proxmox VE, I disabled tablet pointer option, and idle CPU usage went "up".
I then disabled it again and there was no change.
So either the reported CPU usage goes up for the VM simply by being logged into Proxmox UI, or this option has some odd effect that isnt clear. on its impact.
For reference spice is set to none, and I am not logged into the VM, its just sitting idle, and has no desktop package installed.

Confirmed CPU usage went back down after logging out of Proxmox UI.
I now toggled it to disabled, logged out, and idle CPU usage is same as before. So enabled/disabled no change, but VM usage goes up if logged into Proxmox.