Compare

VPS or dedicated server: the question is how often you want to change your mind.

Both give you root on a Linux machine. One is provisioned in about a minute and can be resized, snapshotted and thrown away; the other is cheaper per thread, unmetered on the network, and yours until you migrate off it.

Configure a VPS See the metal line-up

Side by side

Every figure below comes from the same catalogue the cart charges from.

Where we do not have a number — IOPS, port speed on the virtual platform — the row says so instead of guessing.

Cloud VPS compared with a dedicated server across price, tenancy, provisioning, billing, network, storage, resizing, recovery and location
Attribute	Cloud VPS	Dedicated server
Entry price	$7.90 a month for U1-1 — 1 vCPU, 1 GB, no setup fee	$36.90 a month for DS-7 — 8 threads, 32 GB, plus a one-off $38.00
Largest configuration	G224-3840: 224 shared vCPU and 3840 GB at $94608.90 a month	DS-6: 96 threads and 256 GB at $673.90 a month
Tenancy	A virtual machine on a shared host. The Dedicated CPU line pins cores to you; the rest of the host is still shared	The entire physical machine. No hypervisor of ours, no neighbours, no oversubscription
Ready when	Provisioned to order — usually within the hour of payment clearing on the classic lines. The Unmetered and Value lines are resold stock, provisioned to order	Within one business day of the order clearing
Billing	Hourly or monthly, with 10% off for paying twelve months up front. Destroy it and billing stops	Monthly only, plus a one-off setup fee on the first invoice. No hourly rate and no annual discount
Network	Metered on most lines, with a stated allowance from null TB per plan; the Unmetered line has no allowance at all. Port speed is published only where the supplier states one	Unmetered on every machine — 1 Gbit uplinks on the EU builds, 0.5–1 Gbit on the US ones. No allowance to stay under, and no overage conversation
Storage	NVMe on every plan, sized by plan. We do not publish IOPS	Two physical drives — NVMe on every machine except DS-7's SATA pair. EU builds are mirrored at build time if you ask; the US builds arrive as a software mirror. We do not publish IOPS here either
Changing size	Resize up in place from the panel, keeping the IP address. Down needs a rebuild	Not possible. Changing machine is a migration to a different one
Recovery	Free on-demand snapshots, restorable in place or into another region. Nightly backups available at 20% of the plan price	No snapshot and no backup product. A host failure is your own restore, from copies you arranged
Where	48 locations across four areas of the world — 39 regions on the classic lines, a few line-specific sites on the rest	2 countries. The EU builds publish no city; the US builds are in Vint Hill, Virginia
Operating system	The same Linux images on both products. Neither is managed, and there is no Windows licensing on either	The same Linux images on both products. Neither is managed, and there is no Windows licensing on either

The arithmetic

Threads and memory per dollar, using our own prices.

This is the part comparison pages usually leave out, because it does not always flatter the more expensive product.

At the bottom of the range

DS-7 is $36.90 a month for 8 threads and 32 GB. The Value line gets close on the sheet — BD-24 is $30.90 for 8 shared vCPU and 24 GB — but one is a slice of a host and the other is the host. Against the classic lines, metal wins on raw capacity early and keeps winning.

At the top of the range

G224-3840 is $94608.90 a month for 224 shared vCPU and 3840 GB. DS-6 is $673.90 for 96 threads and 256 GB. Per thread, that is not a close contest.

What the difference buys

A rebuild in about a minute, a snapshot before a risky change, a resize from a panel, hourly billing, and — on the classic lines — 39 regions instead of 2 countries. Those are worth real money to a team that is still finding the shape of its system, and nothing at all to one that is not.

Do not forget the setup fee in the sums. DS-7 costs $74.90 in month one and $36.90 thereafter. Over a year that is a rounding error; over the two months you might have spent evaluating it, it is not. The instance has no setup fee and stops billing the moment you destroy it, which is why it is the better place to be uncertain.

Choosing

Two short lists, and no hedging.

Buy the VPS if

Your workload is still changing shape, or you are not certain how big it needs to be.
You want to be running in a minute, not tomorrow.
You need a region outside the 2 countries metal is built in.
Snapshots before risky changes are part of how you work.
You want billing that stops when you destroy the thing.
The system is several small pieces rather than one big one.

Buy the metal if

The load is constant and you already know its size.
You are moving a lot of bytes out and the transfer meter is the binding constraint.
You want cores and memory per dollar above everything else.
Single tenancy is a requirement rather than a preference.
You want to run your own hypervisor and carve the machine up yourself.
A one-business-day delivery and a fixed shape are acceptable.

Questions

The ones that come up after the table.

Is a dedicated server always faster than a VPS?

No, and the cases where it is not are common. A small dedicated machine can be slower than a large instance, and a single-threaded workload on an older server part can be slower than the same work on a high-frequency virtual host. What a dedicated server reliably gives you is exclusivity — nothing else on the machine can take capacity from you — and that turns into predictability rather than raw speed.

At what price does metal start winning?

Compare threads and memory per dollar at the point you are actually at, not in general. Against the classic instance lines, metal gives you far more threads and memory for the money, and the gap widens as you go up — though the Value line narrows it at the bottom of the range. What you give up is the platform: provisioning in about a minute, panel-driven resize, snapshots, and a choice of regions instead of 2 countries. The arithmetic favours metal early; the flexibility keeps favouring the instance for a long time after that.

Can I move from one to the other later?

In one direction fairly easily. Going from an instance to metal means building the new machine, syncing data and cutting over — no different from any other migration, and easier because you can keep the instance running until you are happy. Coming back is the same work in reverse. Neither is a panel button, and neither is dramatic, but budget an evening rather than a coffee break.

Which one gives me better uptime?

Neither, on its own. A single dedicated server is a single point of failure with no live migration and no snapshot to fall back on. A single instance is a single point of failure whose host you do not control. Availability comes from having more than one of something and a tested way to fail over — not from the shape of the box. Buy the shape that fits the workload and design the redundancy separately.

What about a mixture?

Common and sensible: metal for the steady, heavy, bytes-out part of the system, instances for everything that changes shape or needs to exist in another part of the world. Remember that private networking works inside a cloud region only, so traffic between the two crosses the public internet and counts against the instance's transfer allowance where its line has one.

The products

Read the detail.

Other comparisons