Servers & Self-Hosting
11 curated resources, easiest first.
Wiki page covering the Proxmox VE installer: boot the installer, pick a target disk and filesystem (ext4, xfs, ZFS, or btrfs as a tech preview), set the root password and network, then manage it at https://<ip>:8006. The install wipes the selected drives, and ZFS on top of hardware RAID is not supported.
How to install UTM, the virtual machine app for Mac. The Mac App Store version gets automatic updates, while the GitHub DMG is the same build you update by hand. Your VMs are kept when you update, and macOS 11 users are limited to v4.7.5.
Describes VirtualBox network modes: NAT (default, outbound only unless you add port forwarding), bridged, internal, host-only and NAT network. Pick bridged when the VM must be a full peer on your LAN, and NAT when it only needs internet access.
A step-by-step NVIDIA passthrough write-up on Proxmox VE 8.4.9 using Intel VT-d: update BIOS, enable IOMMU in GRUB, create a q35/OVMF VM, blacklist nouveau and nvidia on the host, bind the GPU to vfio and add a hostpci line to the VM. Comments add that vfio_virqfd can be omitted on current releases and that Secure Boot can interfere with guest driver installs.
A tutorial for pulling single files out of a Proxmox VM backup without restoring the whole VM: decompress the .vma.zst, extract the VMA to a raw image, map its partitions with kpartx, mount one and copy files out. The author skipped installing the vma tool on a non-Proxmox machine because its pve-qemu-kvm package has too many dependencies, so run that step on a PVE host.
A solved thread on giving an unprivileged LXC write access to a host folder. The author maps GID 1000 straight through with lxc.idmap, allows it in /etc/subgid, creates a matching datauser group in the container and finally adds root to that group, which was the missing piece.
Solved thread from 2020 where a bind mount showed no usable files because unprivileged IDs are remapped on the host. A Proxmox staff member pointed to the cause, and the author fixed it with subuid/subgid entries and lxc.idmap lines mapping one container ID straight through.
A collection of checks, examples and known issues for passing PCI(e) devices to VMs, KVM only. It tells you to confirm IOMMU and interrupt remapping in dmesg, check IOMMU groups, blacklist host GPU drivers and consider a vBIOS romfile; the ACS override is described as a last resort with risks. A VM with a passed-through device cannot be live migrated.
A community tutorial covering BIOS settings, GRUB or systemd-boot IOMMU parameters, VFIO modules, binding a GPU by PCI ID and verifying with dmesg and lspci. Replies note that amd_iommu=on is redundant on AMD and vfio_virqfd is no longer needed on PVE 8, so check the thread for corrections.
A solved thread about turning an unprivileged LXC into a family Samba server on a ZFS bind mount. The staff answer maps a range of IDs with a single lxc.idmap line, allows it in subuid/subgid and suggests a shared group for ownership; the poster then added each user with smbpasswd and restricted the share to that group.
Explains how container UIDs/GIDs are shifted by 100000 on the host, why bind-mounted files appear as nobody inside the container, and how to fix it with lxc.idmap lines, matching /etc/subuid and /etc/subgid entries, and host ownership changes.