Best Practice Configurations for Virtual Machines on Proxmox VE

Machine Type/Chipset

We generally recommend “q35” as the machine type. This chipset already offers modern PCIe support and is a prerequisite for certain features, such as TPM 2.0 or PCIe passthrough.

BIOS

For new VMs, we use OVMF (UEFI) provided that the guest operating system supports it. SeaBIOS should only be used for legacy systems or older operating system versions, but it is not strictly necessary.

CPU

• The number of sockets should not exceed that of the host.
• The number of vCPUs should not exceed the number of host threads/cores. => In current Proxmox installations, this is no longer possible via the WebUI.
• For the CPU type, we recommend x86-64-v2-AES (or newer) or host.
  • host: While this offers maximum performance, it limits live migration between different CPU generations. Additionally, this setting may cause issues with some operating systems.
  • x86-64-v2-AES (or newer): This option provides a good balance between performance and migration compatibility. => Unless you need all of the host’s CPU flags, this is the recommended configuration. You should select the lowest common denominator as the type across the cluster. This can be done per node and indicates the maximum CPU type supported on that node:
    wget -qO - https://cloud.starline.de/s/QmQt79dpJebtC46/download/checkCPUType.sh | awk -f -
• You should enable NUMA topology if the VM or host has more than one CPU socket. This allows Proxmox to keep a VM’s resources on a single NUMA node.

RAM

• You should define the size in increments of 1024. For example, 8 GB equals 8 × 1024 = 8192 GB, 16 GB equals 16384 GB, and so on.
• Ballooning should always be enabled, but configured with a reasonable minimum value. https://en.wikipedia.org/wiki/Memory_ballooning
• If you experience performance issues, you could try disabling ballooning, though this depends on the workload.
• Huge pages can improve performance for memory-intensive workloads (host-side configuration required).

Storage / Hard Drives

• Controller: VirtIO SCSI Single
• Disk Type: SCSI
• Disk Format: qcow2 (flexible, e.g., for snapshots, even under LVM (Technology Preview) or raw (maximum performance)
  • The supported format depends on the storage. See https://pve.proxmox.com/wiki/Storage
• Enable Discard if the storage backend supports thin provisioning (e.g., ZFS, LVM-Thin, CEPH).
• Enable IO Threads per Disk for better parallel I/O performance.
• Enable SSD emulation if the underlying storage is an SSD or flash system.
• Optional; does not provide any performance improvement and is relevant, for example, if the guest OS expects an SSD.

Network

• Network model: VirtIO (paravirtualized) for optimal performance.
• Leave the MTU set to 1500, unless the network supports jumbo frames and is configured accordingly. For high bandwidths, the multiqueue can be set to a maximum equal to the number of vCPUs assigned to the VM. This enables higher bandwidths.

Sonstiges

• QEMU Guest Agent installieren und in der VM-Konfiguration aktivieren. Ermöglicht sauberere Snapshots, korrekte IP-Anzeige und gesteuerte Shutdowns.
  • Nicht vergessen unter VM > Options aktivieren

Misc

• Install the QEMU Guest Agent and enable it in the VM configuration. This enables cleaner snapshots, correct IP display, and controlled shutdowns.
  • Don't forget to enable it under VM > Options