Proxmox VE 9 vs 8:
what changed

New base: Debian 13 and kernel 6.14

Every major Proxmox release follows a stable Debian version. PVE 8.x was based on Debian 12 “Bookworm”, while PVE 9.0 moves to Debian 13 “Trixie” with the 6.14 kernel. New versions of the key virtualisation and storage components come along with the base.

Snapshots on SAN and RAIDZ expansion

The most impactful practical change concerns storage arrays. Until now, classic SAN setups (“thick” LVM over iSCSI/FC) had no snapshots — often a blocker for VMware migrations. PVE 9 adds snapshots implemented as volume chains, with no need for a clustered filesystem.

• Snapshots on “thick” LVM (SAN / iSCSI / FC) — VM snapshots on block storage, without a clustered filesystem.
• ZFS 2.3 — RAIDZ expansion — add a single disk to an existing RAIDZ vdev without rebuilding the whole pool.
• Faster ZFS dedup and general storage-layer performance improvements.

HA rules: affinity and anti-affinity

PVE 9 introduces a new HA rules system that replaces the old “HA groups”. It lets you precisely control VM placement across the cluster — analogous to the affinity rules known from DRS in vSphere.

• Anti-affinity — keep selected VMs on different hosts (e.g. two nodes of the same app), so a single host failure doesn't take down the whole service.
• Affinity — keep chosen VMs together on one host (e.g. an application close to its database).
• Node affinity — pin a resource to specific nodes; replaces the old HA groups, which are migrated automatically on upgrade.

SDN Fabrics and an nftables firewall

• SDN Fabrics (OpenFabric / OSPF) — build routed networks directly from the UI; very handy for full-mesh topologies for Ceph and cluster traffic.
• nftables firewall — the proxmox-firewall implementation (nftables backend instead of iptables) is mature and the default: faster and more modern.
• Refreshed web UI — UX improvements, a better dark mode and clearer resource views.

Upgrading 8 → 9 safely

Moving from 8.x to 9.0 is an in-place upgrade, node by node — the cluster stays up the whole time. Proxmox ships a dedicated checker script, pve8to9, analogous to the earlier pve7to8.

• First update the cluster to the latest 8.x release.
• Run pve8to9 --full and clear any reported warnings (pre-flight checks).
• Switch the repositories to Debian 13 “Trixie” and upgrade each node in turn, migrating VMs onto it as you go.

If you're only now planning to leave VMware, it's best to start straight on PVE 9 — you get SAN snapshots, HA rules and the latest hardware support without a later upgrade.

What the later 9.x releases brought

The 9.x line evolves quickly, and minor updates add concrete features. The key releases so far:

• OCI images in LXC (9.1) — pull popular OCI images straight from registries and use them as container templates; depending on the image you get a full system container or a lean application container for microservices.
• vCPU flag for nested virtualization (9.1) — precisely enable virtualization extensions in the guest without exposing the full host CPU type (useful for nested hypervisors and Windows with VBS).
• Better SDN visibility (9.1) — the UI shows guests attached to bridges and VNets, EVPN zones report learned IP/MAC addresses, and Fabrics appear in the resource tree.
• Fresher stack (9.1) — Debian 13.2, kernel 6.17.2, QEMU 10.1.2, LXC 6.0.5, ZFS 2.3.4, Ceph Squid 19.2.3.
• Dynamic Load Balancer (9.2) — intelligent cluster balancing: the scheduler factors in real-time node load and automatically migrates HA-managed guests to even out utilisation — the equivalent of vSphere DRS.
• WireGuard and BGP in SDN (9.2) — native WireGuard and BGP, route maps and prefix lists, route redistribution for OSPF and IPv6 support in EVPN.
• Custom CPU models in the UI (9.2) — create, edit and remove custom CPU profiles under Datacenter, without hand-editing config files.
• HA maintenance — disarm/arm (9.2) — temporarily suspend the HA stack for planned work, preserving resource states; base Debian 13.5, kernel 7.0, QEMU 11.0, ZFS 2.4.