When You Outgrow a VPS

A practical guide to deciding whether your next server should be bare metal — without the marketing.

A field-tested decision framework for developers, sysadmins, and small teams running production workloads.

Every VPS reaches a ceiling. Sometimes it's CPU steal time creeping past 5%. Sometimes it's a 4 a.m. OOM kill that you can't reproduce. Sometimes it's the slow realization that you're paying for three VPS instances when one dedicated server would be cheaper, faster, and quieter at night. This page is about recognizing that moment honestly — and choosing the next box without overshooting.

Five signs you've outgrown your VPS

1. 01

   Steal time stays above 5%

   If vmstat 1 shows st consistently above 5, your hypervisor neighbors are eating your CPU. Vertical scaling rarely fixes it.

2. 02

   You've already vertically scaled twice

   You upgraded the plan, then upgraded again. The next tier is bare-metal-priced anyway. At that point, just buy the metal.

3. 03

   Disk I/O is the bottleneck, not CPU

   4k random IOPS under 5,000 on a database-heavy workload is a hint that your storage is shared, undersized, or both.

4. 04

   You're running three or more VPS instances for one app

   The combined monthly cost crosses entry-level dedicated pricing — without the orchestration tax.

5. 05

   You need predictable, sustained throughput

   Game servers, voice/video, ML training, low-latency trading, video transcoding. VPS burst credits won't carry you.

VPS vs Bare Metal — what you're actually paying for

"Bare metal isn't about more — it's about predictable."

The cost crossover

A useful rule of thumb: when you'd otherwise run three or more mid-tier VPS instances for a single application, an entry-level dedicated server usually wins on raw price and operational simplicity. Add the cost of orchestration (load balancer, shared state, network latency between nodes) and the gap widens. Below that threshold, VPS almost always wins on flexibility.

Workloads that almost always belong on bare metal

• High-traffic databases. PostgreSQL or MySQL with hot working sets larger than a typical VPS RAM tier.
• Game servers. Tick rates and player counts that punish noisy neighbors.
• Video transcoding / streaming origin. Sustained CPU and disk pressure.
• ML training and inference. Especially anything that wants GPU passthrough or full PCIe bandwidth.
• Self-hosted SaaS at scale. When you're explaining tail latency to customers, hypervisor overhead becomes a liability.

Dedicated hosting has stayed a focused, performance-driven category — providers like 1Gbits ship dedicated servers and high-spec VPS aimed at exactly these workloads, where shared CPU and capped IOPS silently kill throughput. The point isn't the brand; it's that this category exists for a reason, and it's worth knowing what's in it before you keep stacking VPS instances.

Decision tree

Walk this top to bottom. Stop at the first answer that fits.

Frequently asked

Is bare metal always faster than a VPS?

For sustained CPU, disk I/O, and memory bandwidth: yes. No hypervisor overhead, no noisy neighbors. For bursty or small workloads, a modern KVM VPS can feel indistinguishable in normal conditions.

What's the cheapest case for bare metal over VPS?

When you'd otherwise need three or more mid-tier VPS instances to handle the load. A single entry-level dedicated server is usually cheaper, simpler to operate, and faster end to end.

Do I have to manage the hardware myself?

No. The hosting provider runs the physical machine, network, and IPMI. You manage the operating system and everything above it — the same surface area as a VPS.

Can I migrate a VPS image directly to bare metal?

Not directly. You typically reinstall the OS on the dedicated server and redeploy your apps. Configuration management (Ansible, Terraform, container images) makes this painless if you've invested in it early.