Proxmox Containers vs VMs: When to Use an LXC or a VM

  • Post author:Frank Joseph
  • Post published:
  • Post last modified:
  • Post category:Proxmox
  • Reading time:10 mins read

Proxmox containers vs VMs comes down to the kernel. A container (LXC) shares the Proxmox host’s kernel, so it’s lighter, while a virtual machine runs its own operating system and kernel and is better isolated from the host. For most people, an LXC is the right home for a lightweight Linux service, and a VM is the right home for Docker, Home Assistant OS, Windows, or anything that needs its own kernel.

This was updated for Proxmox VE 9.2, the current release as of October 2026. Since 9.1, Proxmox can also build containers from OCI images, and my own recommendation for Docker has moved from an LXC to a VM.

Proxmox Container vs VM: The Differences

Proxmox VE uses LXC for its containers, managed by its own Proxmox Container Toolkit (pct). A container runs on the host’s kernel instead of emulating hardware, so Proxmox describes the runtime cost as low and usually negligible. The catch is that a container can only run Linux. A VM emulates the hardware and boots its own kernel, which costs more RAM and CPU but runs just about any operating system.

In the GUI, these are the Create CT and Create VM buttons, so Proxmox CT vs VM is the same question as LXC vs VM. Both use the same networking, storage, firewall, backups and high availability.

LXC containerVirtual machine
KernelShares the Proxmox host’s kernelRuns its own kernel
Operating systemsLinux onlyLinux, Windows, BSD and more
OverheadLow, usually negligibleA full OS per VM (more RAM and CPU)
IsolationNamespaces, AppArmor, seccomp and cgroups on a shared kernelBetter isolated, with its own kernel
Migration in a clusterRestart migration onlyLive migration (unless a device is passed through)
SnapshotsDisk state onlyCan include the RAM state
DockerNeeds nesting (and keyctl when unprivileged)Recommended by Proxmox for isolation and live migration
HardwareHost devices by path (dev[n])PCI and USB passthrough, resource mappings

The word “container” trips people up. Proxmox has traditionally positioned its LXCs as system containers, so each one behaves like a small Linux server you log in to and update with apt. Docker runs application containers that each hold a single app. So a generic container vs VM comparison isn’t quite the same question as a Proxmox LXC vs VM.

Unprivileged vs Privileged Containers

An unprivileged container uses user namespaces, so root inside the container is mapped to an unprivileged user on the host. An escape would land as that user rather than root. It’s the default in the GUI, and since Proxmox VE 9.0 (August 2025) it’s the default through the API and CLI too.

A privileged container skips that mapping. The LXC team considers it unsafe and won’t treat new escapes from one as CVE-worthy, so Proxmox’s container (pct) documentation says to use them only in trusted environments.

People usually end up with a privileged container because of storage. Bind mounts into an unprivileged container can hit permission problems from the user mapping. Proxmox VE 9.2 added a per-mount-point idmap option to change the UID/GID mapping, though not on an NFS or SMB share. NFS and SMB mounts inside the container are privileged only: the NFS and SMB/CIFS boxes under Options > Features are greyed out on an unprivileged one. Proxmox also warns that an NFS mount can block the host’s I/O completely. The cleaner pattern is to add the share under Datacenter > Storage and bind-mount it in (pct set <CTID> --mp0 /mnt/pve/<storage>,mp=/mnt/share from the node’s Shell, with your container’s ID and the storage’s ID).

I’ve been on both sides of this. In March 2024, my Jellyfin LXC was privileged so it could use an NFS share, and Docker got its own unprivileged container. As I said then, “you do not want to run Docker with a privileged container. You’re giving root access to your Proxmox server.” Mounting on the host means there’s very little reason to run a privileged container at all.

The container’s Summary shows whether it’s Unprivileged. To convert a privileged one, shut it down, back it up and restore it over itself with Privilege Level set to Unprivileged. Security changes with each release, so treat this as how things stand on 9.2, not a guarantee.

When to Use LXC vs VM

For most people, the rule is simple. If it’s a single Linux service that doesn’t care which kernel it runs on, it goes in an LXC. If it’s Docker, a non-Linux OS, an appliance OS like Home Assistant OS, or something that needs strong isolation or a whole PCI device passed through, it goes in a VM.

In my June 2026 video on moving from a cluster to a single node, my host (Proxmox VE 9.1.16 at the time) kept both: “When I looked at it honestly, only around six important VMs and three important LXCs really needed to stay.” The LXCs were Beszel (server monitoring), Gitea and Twingate, and the VMs were Kasm, Home Assistant, surveillance, Docker, Postgres and Windows 11.

Home Assistant OS needs a VM (see installing Home Assistant on Proxmox), and Pi-hole makes a great first LXC (see installing Pi-hole in an LXC).

Docker, OCI Images and Application Containers

Docker is where my answer has changed, from an LXC to a VM. In an LXC, Docker needs the nesting feature (the Create CT wizard ticks it for unprivileged containers) and, on an unprivileged one, keyctl as well. Proxmox’s documentation recommends nesting containers inside a VM for maximum isolation and live migration.

In March 2024, I ran Docker in an unprivileged LXC with both features turned on. In August 2025, I said “many people including myself have run [Docker in] a Linux container for years without problems.” By February 2026, I’d changed my recommendation: “[The] kernel sharing is the problem. Docker won’t run in a default LXC.” You can still use an LXC, but the host in my June 2026 video ran Docker in a VM. The steps are in my guide on running Docker in Proxmox on a VM.

OCI Images in Proxmox VE 9.1 and Later

Proxmox VE 9.1 (November 2025) added containers created from OCI images, the standard container image format. You pull one with the Pull from OCI Registry button in a storage’s CT Templates view, then create the container with the normal Create CT wizard, and Proxmox converts the image to an LXC. Suitable images make full system containers, but application containers (a single app, like a Docker container) are still a technology preview in the Proxmox VE 9.2 documentation.

As of February 2026, I wasn’t recommending it for Docker yet, because “the traditional process of how you would create and manage Docker containers will be very, very different.” I think it’ll become a stable option at some point, but “for today, I’d recommend using a virtual machine.”

Proxmox VE’s release roadmap plans to mature OCI support and add Compose-based multi-container applications, so it’s one to revisit.

How to Create a Proxmox Container (LXC)

The steps below create a Proxmox LXC from the Debian template on Proxmox VE 9.x. Debian 11 (which this guide used to use) reached the end of its LTS support on August 31, 2026. Use debian-13-standard, the same Debian release Proxmox VE 9 is built on. Some screenshots are from an older release.

1. Log in to Proxmox, select the storage location where you’d like to store the container template (for example, local), select CT Templates, then select Templates.

The CT Templates view of a Proxmox storage with the Templates button at the top.

2. Search for Debian, select debian-13-standard, then select Download.

From the shell, the same download looks like this (the version number changes, so check pveam available first):

pveam update
pveam available --section system
pveam download local debian-13-standard_13.6-1_amd64.tar.zst

3. After the template finishes downloading, select Create CT in the top right to create a new container.

The Create VM and Create CT buttons in the top right of the Proxmox web interface.

4. Enter a Hostname and the root Password (twice), leave Unprivileged container and Nesting checked (both are on by default), then select Next.

The General tab of the Create LXC Container wizard with a hostname, a password, and Unprivileged container and Nesting checked.

5. Select the debian-13-standard template you downloaded, then select Next.

6. Set the Disk size for the container, then select Next.

The Disks tab of the Create LXC Container wizard with the root disk size.

7. Set the number of Cores, then select Next.

The CPU tab of the Create LXC Container wizard with the number of cores.

8. Set the Memory (MiB) and Swap (MiB) (both default to 512), then select Next.

The Memory tab of the Create LXC Container wizard with Memory set to 2048 MiB and Swap set to 512 MiB.

NOTE: This is the limit that this container can use.

9. Set IPv4 and/or IPv6 to DHCP on the Network tab (or enter a static IP), then select Next until you get to Confirm.

The Network tab of the Create LXC Container wizard with DHCP selected for IPv4 and IPv6 on bridge vmbr0.

NOTE: We’re skipping the DNS tab, so the container uses the Proxmox host’s DNS settings unless you change them there.

10. Confirm the settings, then select Finish to create the container.

11. Select the container, then Start and Console, and log in as root with the password from step four. Then run the command below to update the system.

apt update && apt upgrade -y
A container console logged in as root running apt update and apt upgrade.

Managing Proxmox Containers

Containers show up in the resource tree under their node, alongside your VMs, and managing one is no different than managing a VM. Select the container and use the Start, Shutdown and More buttons at the top, or right-click it in the tree.

The toolbar at the top of a Proxmox container with the Start, Shutdown, Migrate, Console and More buttons.

If you have a Proxmox Cluster, you can migrate a container to another node, just not live. Proxmox does a restart migration instead (shut down, move, start again), which it says normally means only some hundreds of milliseconds of downtime.

Container Resource Settings

You can change a container’s CPU and memory later from its Resources tab, and grow its disk with Volume Action > Resize. With Advanced ticked in the CPU dialog, CPU limit (cpulimit) can cap its CPU time (two cores capped at half a core, for example), and CPU units (cpuunits) sets its weight against other containers.

Device Passthrough to a Container

A container can’t use PCI passthrough, but it can get a host device by its path, like a GPU’s render node (/dev/dri/renderD128). This dev[n] option arrived in Proxmox VE 8.1 and reached the GUI in 8.2. Since 9.0, you can add one to a running container, but not edit or remove it.

Only root@pam can configure it. The Resource Mappings under Datacenter are for VMs only, so you give the container the host path directly (the pct.conf reference lists the uid, gid and mode options). Add it under the container’s Resources > Add > Device Passthrough, or run this on the host (101 is the container ID):

pct set 101 --dev0 /dev/dri/renderD128

My Jellyfin guide covers passing the iGPU through to a Jellyfin LXC for hardware transcoding.

Proxmox Container Backup and Restoration

Backups work the same way they do for VMs. You can schedule them or run them manually to an NFS share on a NAS or Proxmox Backup Server, then restore the container from any of them if something goes wrong. That’s how I moved to a single host: “I backed up the important VMs and LXCs from the old cluster and restored them onto the new host.”

A few details are container-specific. Snapshot mode needs snapshot-capable storage, bind mounts and device mount points are never backed up, and an extra volume mount point is only included with its Backup option (backup=1) on. My guide on backing up a VM or container in Proxmox covers the schedule.

Final Thoughts on Proxmox Containers vs VMs

An LXC is a great option when you’re running a Linux service and want a lightweight alternative to a full VM, and it doesn’t hurt to keep one around for testing because of how light LXCs are.

Frank Joseph

I'm Frank, founder of WunderTech. I've been working in enterprise IT for 15+ years and running home labs for nearly a decade — every tutorial on this site is tested on hardware I actually own, including Synology NAS units, a DIY TrueNAS server, a Proxmox cluster, a full UniFi network, and more. I hold a BS in Computer Information Systems and an MBA, but most of what you'll read here comes from my home lab, not a classroom. You can also find video versions of these tutorials on my YouTube channel.