Tutorial, 2026
VPS speed test. Run it for cents, read it correctly.
Benchmarks are easy to run and easy to misread. Here’s what to run, what each number means, and what’s a published fact versus someone’s rule of thumb.
Benchmark mistakes most guides make
- A plain dd writeGoes through the page cacheMisleading
- Geekbench 6 vs 7Different scalesNot comparable
- One runShared hardware variesA sample
Use fio with direct I/O, one Geekbench version, and several runs.
Benchmark a Lite 2
$0.006
- an hour
- YABS
- fio
- iperf3
- Geekbench
- Steal time
A full YABS run costs a few cents. Keep it or delete it.
The short answer
The quickest full VPS benchmark is YABS: one command tests the disk with fio, the network with iperf3 and the CPU with Geekbench. Watch steal time with top as well: it shows how often your virtual CPUs wait for the host.
Read the script before you run it, and skip the parts you don’t want. The reader below explains your numbers.
Most “VPS speed test” pages either test your internet connection or paste a benchmark’s output with no idea what’s good.
This one tells you what each number means, which bands have any published basis, and how to test a server for the price of an hour before you commit to it.
In this guide you’ll learn
- YABS: the one-command benchmark, and its flags
- Disk, network and CPU tests on their own
- Steal time: the number hosts don’t advertise
- A results reader for your numbers
Chapter 1
The one-command benchmark: YABS
Yet-Another-Bench-Script runs three well-known tools and prints the results together.
Read it first
It’s a script from the internet that you’ll run as root. Download it and look before you run it:
curl -sL yabs.sh -o yabs.sh less yabs.sh
Run it
bash yabs.sh
Skip what you don’t need
bash yabs.sh -g # no Geekbench bash yabs.sh -i # no network test bash yabs.sh -f # no disk test bash yabs.sh -n # no IP lookup
By default YABS looks up your server’s IP with an outside service, and Geekbench uploads its result to its own website and prints the link. Use
-nand-gif you’d rather not.
| Part | Tool | What it measures |
|---|---|---|
| Disk | fio, direct I/O, block sizes 4k to 1m | IOPS for small random I/O, throughput for large |
| Network | iperf3, 8 streams, to public servers | Speed to several regions |
| CPU | Geekbench 6 (or 7 with -7) | Single-core and multi-core scores |
Chapter 2
Steal time: the number that matters on a VPS
A benchmark is a snapshot. Steal time shows what happens all day.
One physical server: real CPU cores, shared by everyone on it
KVM hypervisor: schedules every virtual CPU onto a real core
Steal is the share of time your virtual CPUs were ready to run but waited while the host served another machine. Watch the st column:
top # st in the %Cpu(s) line vmstat 5 12 # st column, every 5 seconds for a minute
Check it at a busy hour as well as a quiet one. A one-off spike means little; a high average means the host is crowded.
Chapter 3
What your numbers mean
Enter any of your results. Each line says whether its band is published or a rule of thumb.
What your numbers say
Enter one or more results on the left.
Benchmark a Lite for the price of an hour: a Lite 2 bills by the started hour. Deploy a Lite 2
Chapter 4
Benchmark mistakes to avoid
Easy to make, and they make a fast server look slow or a slow one look fast.
Mythdd tells you how fast the disk is.
A plain dd write goes through the page cache and mostly measures memory. fio with direct I/O measures the disk.
MythGeekbench 6 and 7 scores compare.
They use different scales. Compare a version only with itself.
MythOne run is the answer.
Run at a busy and a quiet hour. Shared hardware varies; a single run is a sample.
Choosing a size? Our VPS sizing guide explains vCPUs and steal in more depth.
Now it’s your move
YABS for the snapshot, steal time for the day-to-day, and the reader for what it all means. Everything else is marketing.
Deploy a Lite, run the benchmark, and decide with real numbers.