Shared vCPU vs Dedicated vCPU: What CPU Steal Is and When It Slows Your Server

Two VPS plans can both say "4 vCPU" and perform very differently. The difference is usually whether those virtual CPUs are shared with other customers or reserved for you. This guide explains how vCPUs work, what CPU steal time is, how to measure it on your own server, and when paying for dedicated cores actually matters.

What a vCPU really is

A vCPU is a slice of a physical CPU thread on the host server, scheduled by the hypervisor. On a shared vCPU plan, the provider places more vCPUs on a host than it has physical threads, assuming not every customer is busy at the same time. On a dedicated vCPU plan, each vCPU you buy maps to capacity that is reserved for you.

What CPU steal time means

Steal time is the share of time your virtual CPU was ready to run but had to wait because the hypervisor gave the physical CPU to another guest. You see it as the st value in Linux tools. A little steal now and then is normal on shared plans. Sustained steal means your server is competing with neighbours for CPU.

How to measure it

vmstat 1 60          # last column 'st'
top                  # '%st' in the CPU line
mpstat -P ALL 5      # per-core %steal (sysstat package)
sar -u 60            # leave running for 24 hours

Measure at different times of day, especially in the evening peak. If steal stays high while your own load is normal, the slowdown is not coming from your application.

Symptoms that point to CPU contention

• Response times that vary at random hours, not with your traffic.

• Benchmarks that give very different results when repeated an hour apart.

• Background jobs, builds or imports that sometimes finish quickly and sometimes take much longer.

• High load average while your processes are not using much CPU.

When shared vCPU is perfectly fine

• Development and staging servers.

• Small websites and blogs with mostly cached pages.

• Bots, monitoring and light automation that do not mind an occasional delay.

• Budget-sensitive projects where the price difference matters more than consistency.

When dedicated vCPU is worth paying for

• Online stores: checkout and cart pages cannot be cached, so every request needs CPU at the moment the customer clicks.

• APIs and SaaS: customers notice inconsistent latency even when the average looks good.

• Databases: heavy queries and index builds suffer when CPU time is taken away mid-task.

• CI runners and build servers: predictable build times make pipelines easier to plan.

• Trading platforms and real-time apps: every delay is visible.

Questions to ask before you buy

1. Are the vCPUs dedicated or shared? Is that written on the plan page?

2. What CPU model runs on the host?

3. Can I move to another node if I see sustained steal?

4. Can I upgrade without reinstalling?

Dedicated resources at IM HOST

Our Cloud VPS plans run on AMD Ryzen/EPYC with fully dedicated vCPU and RAM, from 4 to 24 vCPU, with Windows or Linux and a free automatic backup every 72 hours. If your workload is light and budget comes first, a Linux VPS with full root access and NVMe storage is a good starting point. Either way, renewals stay at your first-term price, and we migrate your current server for free.