Proxmox dedicated servers, root from minute one.

Proxmox VE is an open-source, Debian-based hypervisor that runs both KVM virtual machines and LXC containers from a single web console. On a rented VPS you cannot have it — you are already somebody else's guest. On a Dedizone machine you are the host: your kernel, your firmware, your IOMMU groups, your storage layout. That is the entire difference between a root server with dedicated cores and a dedicated server.

It is also free. Proxmox VE needs no licence key to run; only the enterprise package repository requires a subscription, and the no-subscription repository does not. We charge $0 to install it, exactly as we do for Ubuntu or Debian — the only image on the fleet carrying a licence surcharge is Windows Server, at +$19/mo. Nothing else about the machine changes because you picked a hypervisor: same always-on DDoS mitigation, same unmetered uplink, same 4-hour hardware-replacement SLA.

Consolidation is the reason people run Proxmox on bare metal, so it is worth doing the sums with real numbers instead of adjectives. The DZ-R9 is $115/mo for 16 cores, 32 threads and 128 GB of DDR5 ECC — about $0.90 per gigabyte of memory per month. The DZ-R6 is cheaper still per gigabyte, at roughly $0.77. Scale up and the unit price climbs rather than falls: the DZ-E48 works out at about $1.12 per GB, the DZ-E96 at $1.33.

We would rather state that plainly than pretend the biggest machine is also the best value. What the big machines actually buy is not cheaper memory — it is one failure domain, one address space, one set of disks and one invoice instead of six. And every figure above drops 30% on the 12-month term: the DZ-R9 becomes $80.50/mo, the DZ-E48 $300.30/mo, with $0 setup on every term.

DZ-R9 — Ryzen 9 9950X, 16c/32t, 128 GB DDR5 ECC, 2× 2 TB NVMe Gen4, unmetered gigabit, $115/mo. The right first Proxmox host for most people: high clocks make single-threaded guests feel fast, and a mirrored pair of 2 TB NVMe holds a few dozen workloads comfortably.

DZ-E48 — EPYC 9454P, 48c/96t, 384 GB DDR5 ECC, 2× 3.84 TB NVMe U.2, 10 Gbit/s, $429/mo. Three times the memory of the DZ-R9 on enterprise U.2 drives. This is the machine our own fleet notes list for virtualisation, databases and heavy compute, and it is the sensible ceiling for a single-node private cloud.

DZ-E96 — dual EPYC 9654, 192c/384t, 1.5 TB DDR5 ECC, 4× 7.68 TB NVMe U.2 (30 TB raw), 25 Gbit/s, $2,049/mo. A datacentre in a box. It earns its price only when a rack of smaller nodes would cost you more in coordination than in money. If a guest needs a GPU instead, the DZ-G1 and DZ-G2 put one or two RTX 5090s on a PCIe bus you control.

There is no honest benchmark answer to this, because it depends entirely on what the guests do. There is an honest arithmetic answer, and that is the one worth publishing. Reserve 16 GB for the host and the ZFS cache on a DZ-R9 and 112 GB remain: twenty-eight guests at 4 GB, or fourteen at 8 GB. On a DZ-E48, reserving 32 GB leaves 352 GB — about eighty-eight 4 GB guests. On a DZ-E96, roughly 350.

Threads are the softer limit. Proxmox will let you allocate more vCPUs than you have threads, and for idle web guests that overcommit is free; for anything CPU-bound it is a promise you cannot keep. Count memory first, because memory is the resource you cannot overcommit without a swap-death spiral, then check that the handful of genuinely busy guests still fit inside the real thread count.

Disk layout is a build choice made at order time, not a paid add-on, and it is the one decision that is genuinely awkward to undo later. Three options, all free: RAID 1 mirrors the pair, which is the default and what most people want; RAID 0 stripes for capacity and throughput at the cost of all redundancy; JBOD hands you the raw devices with no array in the way.

Pick JBOD if you intend to run ZFS. The Proxmox installer will build a mirrored ZFS root across raw disks itself, and ZFS wants to see physical devices rather than a hardware array — that is where snapshots, native replication and cheap thin clones come from. On a DZ-R9 that means a 2 TB mirror or 4 TB striped; on a DZ-E48, 3.84 TB mirrored; on a DZ-E96, four 7.68 TB drives to arrange as you like. The same reasoning drives our 216 TB storage server, which ships its twelve SAS drives as raw disks for exactly this purpose.

Every server comes with one IPv4 address and a /64 IPv6 block, plus reverse DNS control. For a Proxmox host that shapes the design: the IPv4 sits on the host, guests hang off a Linux bridge, and you either NAT them outbound or give them addresses from the /64 — which is far more public addressing than any single box can exhaust.

If specific guests need their own public IPv4 — a mail server, a TLS endpoint that cannot rely on SNI, a licence-bound appliance — extra addresses are +$3/mo each and the cap is four. That is a real limit, so we will state it plainly: five public IPv4 addresses per server is the ceiling, and past that the answer is IPv6, a reverse proxy, or a second machine. Traffic between the host and its own guests never touches the uplink at all, and the uplink is unmetered at the listed speed, so east-west chatter costs nothing.

There is no ISO to upload, no rescue-mode bootstrap and no ticket to open. On the order form you pick the machine, one of eight European locations, Proxmox VE 9 from the OS list, the disk layout, a root password or an SSH key, and any extra addresses. Then you pay a single-use invoice in Bitcoin, Monero, USDT, Ethereum, Solana, Litecoin or Tron.

When the payment confirms on-chain the pipeline images the machine to that spec — hypervisor, disk layout, network, reverse DNS — and the credentials arrive by email. The SLA commits to online within five minutes, typically under sixty seconds, or the first month is half price; if we somehow fail to deliver inside 24 hours you may take a full refund in the asset you paid. The same automation covers a rebuild, so reinstalling a host you have made a mess of costs nothing but the time.

The Proxmox web console answers over HTTPS with a self-signed certificate, which is worth replacing before you do anything else. After that the ordinary sequence is short: point the package manager at the no-subscription repository, update, confirm the bridge picked up your address, then create the first storage — a ZFS pool if you took JBOD, LVM-thin if you took RAID 1.

The first KVM guest wants a virtio disk, a virtio NIC and the guest agent installed; the first LXC container wants a template and about ninety seconds of your life. Snapshot both before you touch them again. None of this is Dedizone-specific, and that is rather the point: you are running upstream Proxmox on hardware you control, not a vendor panel that will be deprecated in two years and take your automation with it.

Snapshots are not backups. A snapshot lives on the same pool as the thing it is protecting, which means it survives your mistakes but not the pool. Proxmox Backup Server solves that properly with deduplicated, incremental, verifiable backups — and it needs somewhere to live that is not the host it protects.

Two honest options. The 500 GB off-server backup add-on is +$9/mo of BorgBackup or FTP space in a different data centre: enough for configuration, databases and dump archives of small guests, and not enough for a fleet of full disk images. Be clear-eyed about that before you rely on it. For real capacity, run Proxmox Backup Server yourself on a second box — a DZ-S216 gives you 216 TB of raw SAS at roughly $2.08 per terabyte per month, with unmetered bandwidth so a full restore never becomes an egress bill. We do not operate PBS as a managed service, and we would rather say so now than have you discover it mid-restore.

Storage replication between Proxmox nodes is the cheapest disaster recovery there is: pick a dataset, pick an interval, and the second node holds a recent copy. Two Dedizone servers in two jurisdictions — Frankfurt and Amsterdam, say, or Zurich and Helsinki — give you that for the price of the second machine and nothing more, because bandwidth is unmetered at both ends.

What two nodes do not give you is automatic high availability. Proxmox HA needs quorum, and quorum needs an odd number of votes; a two-node cluster will fence itself the first time the link between the cities blinks, unless you add a third vote. So: two nodes for replication and a deliberate human failover — cheap, robust and honest. Three nodes if you want the cluster to make that decision for you. Anyone describing a pair of boxes as an HA cluster is selling you something.

Hypervisors tend to fail in ways that lock you out of the machine rather than crash it outright: a kernel update that will not boot, a bridge change that takes the management interface with it, a storage pool that wants a decision at the console. Every Dedizone server includes permanent IPMI/KVM-over-IP with virtual media, from the first minute, at no charge.

That means a browser console with keyboard, video and mouse, power control, and the ability to mount an ISO — the difference between a ten-minute fix and a lost evening. It is also why we keep making the point: Hetzner lends a KVM console for three hours, Netcup offers a VNC window into a VM you do not own, and Kimsufi gives you rescue mode and good luck. For a machine whose entire job is hosting other machines, out-of-band access is not a feature — it is the floor.

A Proxmox host is a larger target than a single application server, because compromising it compromises everything on it. The short list: do not expose the web console to the open internet — bind it to a management address or reach it over a WireGuard tunnel; use the built-in firewall at datacentre, node and guest level rather than only inside the guests; keep an SSH key and disable password login; and keep the guest bridge away from anything that can talk to the host.

Always-on L3/L4 DDoS mitigation runs upstream on every server across more than a terabit per second of scrubbing capacity, so volumetric attacks aimed at an exposed guest are absorbed before your uplink notices — and because bandwidth is unmetered, filtered attack traffic never turns into an invoice. What we cannot do is see inside your guests. We install no agents, we run no management plane on the box, and there is no managed tier waiting to be upsold to you.

Crypto only: Bitcoin, Monero, USDT on TRC-20 or ERC-20, Ethereum, Solana, Litecoin and Tron. The invoice is single-use, the amount is locked for the payment window, and if you underpay the page shows the exact remainder to send to the same address. Identity is an email address — no documents, no phone, no company registry. If that is the part you came for, the no-KYC and Monero pages go into it properly.

Terms are one, three or twelve months, at full price, minus 10% and minus 30% respectively, with $0 setup on all three and no minimum commitment. Renewals are invoiced the same way and each one may be paid in a different asset. Weighing us against Contabo or anyone else on the market, the number worth checking is not the headline rate but what the setup fee and the commitment length do to your first year.