Comparison with our own test, 2026
VPS vs dedicated server: which one do you actually need?
The real differences, a load test on our own server, and a 30-second check that tells you which plan fits.
VPS (shared vCPU)
Your vCPUs take turns on cores other customers use too.
Dedicated vCPU
Some cores are reserved for you. The rest of the machine is shared.
Dedicated server
Every core, disk and network port is yours, and so is every fault.
Yours Another customer Shared, idle
The short answer
A VPS is a virtual machine on hardware shared with other customers. A dedicated server is a whole physical machine rented to you alone. The VPS starts cheaper, bills by the hour and is usually ready within 15 minutes. The dedicated server usually means a monthly contract, a setup fee, and every hardware fault is yours to handle.
For almost every project, a big VPS is the better buy. On LiteKeep, Lites go up to 64 vCPU and 256 GB of RAM on KVM with NVMe, billed by the hour with no setup fee. The free check below tells you which size you need.
Most pages on this question say the same thing: a VPS is cheaper, a dedicated server is faster, it depends.
We went further. We kept every core of our own VPS busy for a minute to see how much CPU the host took back, and turned the warning signs into a free check you can run on your own server.
The result: under full load, our host took back 0.15% of the CPU. For most people, that settles it.
In this guide you’ll learn
- The difference in one table: what you rent, billing, upgrades, a dead disk
- What each really costs, including what the monthly price leaves out
- Noisy neighbours, measured: our VPS under full load
- A free check that ends in the plan to deploy
- The reasons people go dedicated, and what a big VPS does instead
Chapter 1
The difference in one table
Log in to either one and you see the same thing: a Linux server with root access. The difference is what sits under it, and who deals with it when it breaks.
- A VPS is a virtual machine. On KVM, each VM is a process scheduled by the host’s Linux kernel, with its own virtual hardware and its own kernel. Your vCPUs take turns on the host’s real cores.
- A dedicated server is the physical machine: its CPU, RAM, disks and network port, rented to one customer.
| VPS on LiteKeep | Dedicated server | |
|---|---|---|
| What you rent | A KVM virtual machine | The whole machine |
| Ready to use | Usually within 15 minutes | Often hours or days |
| Billing | By the hour, capped at the monthly price | Monthly, often with a setup fee |
| Need more power | Deploy a bigger Lite and move over | Order a bigger machine and migrate |
| A disk dies | Handled for you | You spot it, book a swap, rebuild the RAID |
| Locations | 15 countries, one price | Wherever that provider has stock |
| How you pay | Crypto, no card, no ID | Usually a card and a contract |
The disk row is not a figure of speech. On a dedicated server, a failed disk means reading /proc/mdstat yourself, booking a swap with support, taking the server offline and rebuilding the array afterwards. On a VPS you never see it.
Chapter 2
What it really costs
Dedicated servers often look good per core on a price list. The bill tells a different story once you add what the list leaves out.
| Plan | Size | Monthly |
|---|---|---|
| Lite 64 | 24 vCPU, 64 GB RAM, 1,280 GB NVMe | $80 |
| Lite 128 | 40 vCPU, 128 GB RAM, 2,560 GB NVMe | $160 |
| Lite 256 | 64 vCPU, 256 GB RAM, 5,120 GB NVMe | $320 |
What a dedicated server’s monthly price leaves out
- The setup fee. Often a month’s rent or more, paid before you’ve run a single job. A Lite has none.
- Your time. Disk failures, RAID rebuilds and hardware swaps are your job on a dedicated server.
- Commitment. A setup fee only pays off if you keep the machine for months. Hourly billing lets you test a 64-vCPU Lite for an afternoon.
- Spare capacity. You buy a machine for your peak and pay for it every quiet day. With hourly billing, you can run a big Lite only while you need it.
Chapter 3
Noisy neighbours: our test
The usual case for dedicated hardware is the noisy neighbour: another customer’s busy VM slows yours. Linux measures that directly. The kernel counts steal time: the time your vCPU wanted to run but the host gave the core to someone else.
Here is what we did. On 24 October 2026 we kept all 4 vCPUs of our own KVM server at 100% for 60 seconds, using plain shell loops, and read the CPU counters every 20 seconds.
| Measured | CPU busy | Steal |
|---|---|---|
| Since boot (almost 5 days, light load) | Mostly idle | 0.04% |
| Full load, seconds 0 to 20 | 99.7% | 0.18% |
| Full load, seconds 20 to 40 | 99.8% | 0.14% |
| Full load, seconds 40 to 60 | 96.2% | 0.15% |
| Full load, whole minute | 98.6% | 0.15% |
Under full load the host took back about one and a half thousandths of our CPU time. There was no noisy neighbour to escape.
Want to see your own number? The check below samples your server at your busiest hour.
Chapter 4
Have you outgrown your VPS? Check it
Run this block over SSH while your server is at its busiest and paste the output. You’ll get a verdict and the LiteKeep plan that fits.
s() { awk '/^cpu /{print $2+$3+$4+$7+$8, $5, $6, $9}' /proc/stat; }
a=$(s); sleep 30; b=$(s)
echo "$a $b" | awk '{b=$5-$1; i=$6-$2; w=$7-$3; s=$8-$4; t=b+i+w+s; printf "busy=%.1f\niowait=%.1f\nsteal=%.1f\n", 100*b/t, 100*w/t, 100*s/t}'
echo "vcpus=$(nproc)"
echo "load15=$(cut -d' ' -f3 /proc/loadavg)"
awk '/^MemTotal:/{t=$2} /^MemAvailable:/{a=$2} /^SwapTotal:/{s=$2} /^SwapFree:/{f=$2} END{printf "mem_total_mb=%d\nmem_avail_mb=%d\nswap_used_mb=%d\n", t/1024, a/1024, (s-f)/1024}' /proc/meminfo
echo "oom_kills=$(awk '/^oom_kill /{print $2}' /proc/vmstat)"
echo "disk_used_pct=$(df -P / | awk 'NR==2{sub("%","",$5); print $5}')"Read-only, no sudo. Takes 30 seconds. Works in sh and bash.
Run the block, then paste its output on the left.
Our thresholds: CPU busy 85% or more during the sample, a 15-minute load above 1.5 per vCPU, steal of 10% or more, disk wait of 10% or more, under 10% of RAM available, any out-of-memory kill, or a root disk 90% full. Judge trends over a few busy hours, not one sample.
Most red results point to the next size up, not new hardware. Running out of RAM on a 4 GB plan is a 4 GB problem; a Lite 6 or Lite 8 fixes it the same day.
Chapter 5
Why people go dedicated, and what a big VPS does instead
These are the reasons people give for buying a dedicated server. For most of them, a large Lite does the job with less cost and less work.
“I need more power”
Lites go up to 64 vCPU and 256 GB of RAM with 5 TB of NVMe. Most sites, APIs, databases and game servers never get near that.
“I don’t want neighbours slowing me down”
Measure it. On our KVM test, steal stayed at 0.15% under full load. If your current host shows double digits, the fix is a quieter host, not new hardware.
“A dedicated server is more secure”
It removes the neighbours. Weak passwords, open ports and missing updates are just as dangerous on both, and that’s where attacks really start. Every LiteKeep server is a full KVM machine with DDoS filtering included. Our VPS security guide covers the rest.
“It’s cheaper at my size”
Only once you ignore the setup fee, the contract and your own time on hardware. Hourly billing means you pay for a big Lite only while it runs. See what a VPS really costs.
Now it’s your move
A VPS rents you a share of a machine, billed by the hour and usually ready within 15 minutes. A dedicated server rents you the machine, with a setup fee, a contract and every hardware fault yours.
Run the check at your busiest hour, then deploy the Lite it points to: KVM, NVMe, DDoS filtering, 15 countries, paid in crypto. Our guide to choosing a VPS covers the rest, and our KVM explainer shows what runs under it.