Dedicated Servers: Bare Metal Power for Demanding Workloads

A dedicated server is one physical machine, allocated to one customer, with no hypervisor between your operating system and the hardware. Every core, every memory channel, every megabyte of L3 cache and every disk bay belongs to a single workload. Our range starts at €209/month for 16 physical cores and 128 GB of DDR4 ECC memory, with IPMI from day one and provisioning inside an hour.

What you are buying that a VPS cannot give you

The headline difference is not core count — you can buy a lot of vCPU. It is that four resources stop being shared:

• Physical cores. No hypervisor scheduling, no steal time, no other guest ready to run when your thread is.
• Memory bandwidth. Our dual-socket machines expose eight DDR4 channels, four per socket. Memory-bound workloads scale with channels, and on a shared host you get a fraction of them.
• L3 cache. A neighbour cannot evict your cache lines, because there is no neighbour. For workloads with a working set near cache size, this alone can be the difference.
• Disk. Physical bays — six on the Xeon E5 range — in a RAID level you choose, on a controller you control.

ECC memory is the fifth, quieter benefit. Single-bit errors are corrected rather than silently corrupting a long-running process, which matters for databases, in-memory caches and anything running for months at a time.

The range

Machine	Cores / threads	Memory	From	Suits
Dual Xeon E5-2630 v3	16c / 32t	128 GB DDR4 ECC	€209/mo	Container hosts, CI runners, web tiers
Dual Xeon E5-2690 v4	Broadwell-EP	DDR4 ECC	—	Higher clock than the v3, mixed workloads
Dual Xeon E5-2699 v4	High core count	DDR4 ECC	—	Maximum thread count on the Intel side
EPYC 9254	Modern Zen	DDR5 ECC	—	Strong per-core performance
Dual EPYC 9254 · 9554 · 9754	Up to very high density	DDR5 ECC	—	Virtualisation hosts, heavy parallel compute

Full specifications, pricing and workload guidance for each machine are on the dedicated server range. Every one ships on a 1 Gbps guaranteed unmetered port with no traffic cap and no overage billing, in any of our twelve locations.

Workloads that justify bare metal

Virtualisation hosts

If you want to run Proxmox, plain KVM or VMware and decide your own overcommit ratio, you need the physical machine. IPMI lets you mount your own ISO, so the hypervisor choice is yours rather than ours.

Databases with latency budgets

Average query time is usually fine on a VPS. The p99 is what moves when cache and memory bandwidth are contended. A database with a strict tail-latency requirement is one of the clearest cases for dedicated hardware — see our database server page.

Transcoding and media processing

Video encoding scales almost linearly with physical cores and is memory-bandwidth hungry. It is the archetypal bare-metal workload, and it usually wants a large port as well — see transcoding and media processing.

Build farms and CI at scale

Compilation is embarrassingly parallel and starves on memory bandwidth long before it starves on clock. Thirty-two threads across eight channels is a very different machine from thirty-two vCPU on a shared host.

Compliance-driven single tenancy

Some contracts and some regulatory regimes require that no other tenant occupies the physical hardware. That is binary — no degree of VPS isolation satisfies it.

What comes with the machine

• IPMI from day one. Remote console, power cycling and OS reinstall without opening a ticket. This is the difference between a two-minute recovery and a two-hour one.
• Your operating system. Ubuntu, Debian, AlmaLinux, Rocky and CentOS at deploy; anything else via custom ISO, including Windows Server with your own licence.
• Provisioning inside an hour. Not instant — physical hardware never is — but same-session rather than next-day.
• Unmetered 1 Gbps. Saturate it continuously without a surcharge.
• DDoS protection in front of the network, not sold as an add-on.

Where a dedicated server is the wrong purchase

Three cases, stated plainly:

• Your load is spiky. A physical machine bills the same at 5% and 95% utilisation. If your peak is four times your median, you are buying idle metal — hourly VPS billing is the cheaper instrument.
• Your workload is single-threaded. Sixteen cores at 2.4 GHz will not run a single-thread-bound process faster than four cores at 3.5 GHz. Match the machine to the shape of the work.
• You want redundancy more than power. One machine is one failure domain. Three VPS instances across hosts survive a fault that takes a dedicated server offline. Availability is architecture, not hardware.

VPS versus dedicated servers works through the crossover in detail, including the price arithmetic.

Sizing without guessing

1. Profile the real workload. Peak resident memory, sustained core utilisation, disk IOPS, and whether the hot path is parallel or serial.
2. Pick clock or cores, not both. Serial work wants clock; parallel work wants cores and memory channels. Very few budgets buy the maximum of each.
3. Size memory for the peak, not the average. Adding RAM to a physical machine means a maintenance window; getting it right at order time does not.
4. Plan the array before handover. Tell us the capacity and RAID level you want and it is configured before you get the machine — RAID 1, 10, 5 and 6 are all supported on the onboard controller.
5. Benchmark on an hourly VPS first. A €0.0806/hr Titan instance for a day costs €1.93 and tells you far more about your scaling behaviour than a spec sheet does.

Frequently asked questions

How long does provisioning take?

Under an hour for a standard configuration. Custom RAID or unusual disk capacities add time because the array is built before handover.

Can I install my own hypervisor or Windows Server?

Yes. IPMI lets you mount any ISO, so Proxmox, VMware and Windows Server all work. Windows requires your own licence.

Is bandwidth genuinely unmetered?

Yes — 1 Gbps guaranteed, no traffic cap, no overage billing. For sustained multi-gigabit throughput, the 10 Gbps streaming range is the cheaper route than a dedicated machine on a 1 Gbps port.

What happens if hardware fails?

Faulty components are replaced. That is a physical operation with real downtime, which is why anything with a strict availability target should be designed across more than one machine rather than relying on one being reliable.

Do you offer management?

Dedicated machines are unmanaged by default — you have root and IPMI, and the operating system is yours. Managed versus unmanaged hosting covers what that means in practice and how much time it realistically costs.

Getting started

Pick the machine on the dedicated range, choose a location close to your users, and tell us the disk configuration you need. If you are not yet certain bare metal is the right answer, run the workload on a large hourly VPS for a week first — the measurement costs a few euros and removes the guesswork entirely.