VM's start to a black screen unless I allocate less than 1/4 of available cores

F

futiless

Renowned Member
Jul 4, 2017
53
19
73
45
North Vancouver, British Columbia : Canada
There seems to be no indication as to the best way to launch my VM's with all the core's in the system in this case 36 are advertised on the Node System summary panel. This is comprised of 1 socket, 18 Core (E5-2699 v3) and 36 with hyper threading. It doesn't make sense when I have only 4 vm's with passthrough graphics to limit them to less cores than available. After all it would seem as though the best way to load balance the processor is with CPU units greater than the default 1024.

PLEASE help me get this computer working I would do back flips of joy!! :)
 
Maybe some NUMA problem. Please try to enable NUMA in the CPU settings dialog.

In general, it is not a problem to start a server with all available CPU cores (or even more).
 
  • Like
Reactions: futiless
Fair enough; but is there a way to understand this logic, or try and comprehend how the computer is canibolizing the cores. In my case the best EVER I have seen was 14 cores... ( OF 36 ?_)) I need someone to explain throughly to me how the cpu setting allocation on the VM works the manual is missing piles of information. (with NUMA, and host as cpu type and 1025 units.) If I switch to Haswell (BTW: I have a 18 core/36thread, which the node identifies as 36 on the summery panel. I mean once in a while after changing a pile of settings I can get it to boot with 18 but never as host. And it won't work twice.. just an off chance. But that is a huge loss of resources if I want to use PROXMOX longterm and continue to use it I have to somehow alleviate the concern that resources are going underutilizes to a degree in which inspires tampering and longterm messing around with these settings.
 
Last edited:
Maybe the limitation is due to the GPU passthrough? Have you tried to create a VM without a GPU passthrough and test if that works fine?
 
futiless said:
Assuming that solves the problem how would this be resolved.
Click to expand...

Don't do passthrough. The command to do this not to do anything that involves passthrough, which has to be done manually.

futiless said:
Do you know various commands I could use to assess what is going on. ?
Click to expand...

If this is a "real" QEMU problem, you have to look into the QEMU monitor. There is a help page that has various commands to live debug the virtual CPU but I think this is for CPU geeks only. I wouldn't know how to debug the memory on the x86_64 architecture.
 
  • Like
Reactions: futiless
Are you planning on building a rig for gaming or why do you want to passthrough all the GPUs? A lot of the great features of virtualization cannot be used if passthrough is used.
 
You must log in or register to reply here.
Top Bottom