Choose VPS if
- Traffic varies and you want to grow tier by tier.
- The application fits comfortably under 32 vCPU and 192 GB RAM.
- You want a quick snapshot before every risky update.
- You run several small environments: test, staging and production.
A KVM VPS gives you reserved resources on a physical host shared with other customers. A dedicated server gives you the whole machine: CPU, memory, disks and network card. These are the criteria that decide when to switch: steady performance, isolation, storage, compliance, delivery time and scaling.
| Criterion | VPS | Dedicated server |
|---|---|---|
| Hardware | Reserved resources on a shared host | A whole physical server, yours alone |
| Maximum 4hosting capacity | 32 vCPU, 192 GB RAM | 48 cores, 256 GB RAM |
| Peak performance | Can vary because of neighbours | Constant, no neighbours |
| Isolation | KVM virtualisation, own kernel | Complete physical isolation |
| Storage and IOPS | Virtualised NVMe, reserved volume | Dedicated NVMe, disks in pairs |
| Scaling | Up, to a larger tier | Hardware upgrade or migration |
| Snapshots | Included | No hypervisor snapshots, classic backup |
| Delivery time | Fast, a virtual instance | Longer, installed on hardware |
| SLA | No 99.99 % SLA | 99.99 % SLA |
| Included DevOps hours | 1–8 h per month | 5–10 h per month |
A VPS reserves CPU, memory and disk for you, yet the physical host underneath is shared. KVM, the virtualisation layer built into the Linux kernel, gives each VPS its own kernel and its own memory. What stays common is the processor cache, the memory bus and the disk I/O queue. When another tenant runs a heavy job, your response times can spread out during peaks. A typical website or API will hardly show it. A database handling thousands of writes per second will. On a dedicated server the question disappears, because no other workload runs on that machine.
All 4hosting dedicated servers use NVMe disks in pairs. The SD Intel 32/256 plan, with a 32-core Xeon 6731P, has four 3.84 TB disks, and you pick the RAID level when the server is installed: RAID 1 on two disks or RAID 10 on four. A VPS gets a virtualised NVMe volume reserved for it. You feel the difference under many small random writes, which is how transactional databases and message queues behave. A dedicated server hands you the full disk bandwidth, while a VPS volume is plenty for most web workloads.
If you handle patient, payroll or payment data, an auditor will ask where it is stored and who else can reach the hardware. Both 4hosting options run in an EU data centre in Frankfurt, covered by a data processing agreement based on Article 28 GDPR. A dedicated server goes one step further with physical isolation: no other customer shares your disks. A well-configured VPS satisfies many audits. When your own policy or a large customer's contract names dedicated hardware outright, you already have your answer.
A shop that peaks on Black Friday and idles in summer begins on a 4 vCPU VPS and moves up a tier before each campaign. It takes a snapshot ahead of every platform update, which gives a fast rollback when an update misbehaves.
A B2B SaaS grows a PostgreSQL database to several hundred GB, with writes all day. On a large VPS the query times turn uneven at peak hours. Putting the database on a dedicated server with NVMe in RAID 1 evens them out, and the application stays on the VPS.
A clinic group keeps patient records, and an auditor asks that the data be physically isolated. The group takes a dedicated server in an EU data centre, picks the RAID layout at install, limits root access to its own team and signs a data processing agreement based on Article 28 GDPR.
Yes, if the application runs in Docker containers or its configuration is described as code. The move then comes down to starting the same containers on new hardware and switching DNS. Without containers, migration means copying data and reconfiguring services, work the 4hosting team can do from the included DevOps hours.
There is no fixed ratio. A vCPU usually maps to one hardware thread of the host, not to a whole core. The only valid test is your own workload: run the application on both options and compare response times at peak load.
Start on a VPS while load varies and you grow in steps; move to a dedicated server when you need the whole machine, steady IOPS and physical isolation.