VPS Compare: How to Read the Specs Before the Price, and When the Comparison Is the Wrong One

Comparing VPS plans by price and RAM is how people end up with a cheap server that is slow. The number that matters is what the vCPU is, shared or dedicated, and what happens when the neighbours are busy; then the disk type and the bandwidth cap; then the price. And for a lot of the people running the comparison, a VPS is not the right column at all, because the cost they are about to pay is measured in evenings, not dollars.

The sites that rank for this search are comparison tables: hundreds of plans from dozens of providers, sortable by price, RAM, vCPU and location. They are genuinely useful once you know what you are looking for and dangerous before, because a table makes every column look equally important and lets you sort by the one that is easiest to understand. This post is about how to read those tables, and about the question the tables cannot ask.

Start with the workload, not the table
What the spec columns actually mean
The line items that are not in the table
A five-minute benchmark that beats the table
Managed vs unmanaged is the bigger decision
How a managed plan compares on the same axes
How this fits the rest of the stack
FAQ

Start with the workload, not the table

A comparison without a workload is a shopping trip without a list. Before opening a table, write down three things: what runs on the server, how much memory it needs at peak, and whether the CPU is busy all day or a few seconds at a time.

A WordPress site with a caching plugin idles at a few hundred megabytes and spikes when someone rebuilds the cache. A Node API with a connection pool uses a steady half a gigabyte and wants a fast single core. A Postgres database wants memory above everything and a disk that does not stall. A build server wants many cores for ten minutes and nothing for the rest of the hour. These four workloads sort a comparison table four different ways, and the plan that wins for one loses for another.

The 16GB VPS post has the method for working out how much memory you actually need. Do that arithmetic first. Then the table has a row range instead of a hundred rows.

What the spec columns actually mean

Every table has the same five columns, and every one of them hides something.

vCPU. A vCPU is a slice of a physical core’s time, and the plan page rarely says how big a slice. Shared vCPUs are oversold, which is fine for bursty workloads and bad for anything that computes continuously; a provider may throttle a shared core that stays busy. Dedicated vCPUs are a reserved core or hyperthread. Two shared vCPUs can be slower than one dedicated one under sustained load. The high performance VPS post covers how to verify the claim rather than trust the label.

RAM. The one honest number, with one catch: some plans count swap or count a burstable allocation. Look for the word guaranteed.

Storage. SSD and NVMe are not the same. NVMe is several times faster for random reads, which is what a database does, and a plan that says SSD in 2026 may be older hardware. The size matters less than the type for most apps; the VPS SSD post goes into why.

Bandwidth. Read this as two numbers: the monthly transfer cap and the port speed. A 1TB cap on a 100Mbps port is a different plan from a 1TB cap on a 1Gbps port, and the overage rate after the cap is the number that appears on the surprise invoice.

Location. Latency to your users is the only reason to care, and it is worth more than any other column for a site whose visitors are in one country. Fifty milliseconds of round trip is the difference between a fast page and an average one, and no amount of CPU buys it back.

The line items that are not in the table

The tables compare the plan price. The bill has more lines than that, and they are where a cheap plan becomes an expensive one.

Backups. Usually a percentage of the plan price for nightly snapshots. Without them, the disk is the only copy of your data.
A static IP or extra IPs. Sometimes included, sometimes a monthly charge, occasionally a one-off.
Bandwidth overage. Per gigabyte after the cap; find the rate before the cap is hit.
Managed support. An unmanaged VPS means you are the operator. A managed tier adds a monthly fee and moves patching, monitoring and some troubleshooting to the provider.
The control panel licence. cPanel and Plesk are licensed per server or per account, and the price has climbed every year.
Your time. Updates, a firewall, a reverse proxy, certificates, a process manager, log rotation, the reboot after a kernel patch. Two hours a month at any reasonable rate exceeds the plan price of most small VPSs.

The VPS price post covers what drives the headline number. The point here is that the headline is not the total, and the tables cannot show the total because the largest line depends on who you are.

A five-minute benchmark that beats the table

Most providers offer hourly billing or a short refund window. Use it. Rent the plan for an hour, run three commands, and you will know more than the comparison site does.

# CPU: single-core throughput, roughly comparable across providers
sysbench cpu --cpu-max-prime=20000 --threads=1 run | grep "events per second"

# Disk: random 4k reads, the pattern a database produces
fio --name=randread --rw=randread --bs=4k --size=1G --numjobs=1     --iodepth=32 --runtime=30 --time_based --direct=1 --filename=/tmp/fio.test

# Steal time: how much the hypervisor is taking from you
vmstat 5 6

Read the st column in vmstat. A steal figure that stays above a few percent means the host is oversold and the vCPU number on the plan page is aspirational. Run the disk test twice, an hour apart; a large difference means you are sharing the disk with someone busy.

Then run your actual application for the rest of the hour, with a load-testing tool pointed at it from another machine, and watch memory. This is the only test that tells you whether the plan is big enough, and it costs a fraction of a cent.

Managed vs unmanaged is the bigger decision

Comparison tables treat every VPS as the same kind of product with different numbers. The larger split is what you get with the server, and it is the split most people should decide first.

An unmanaged VPS is a machine with an operating system. Everything above that is yours: the web server, the runtime, TLS, the database, backups, monitoring, security patches. It is the cheapest column in every table and the most expensive in hours. Choose it when the server is the point, you want to learn, or the workload genuinely needs root and a custom kernel module.

A managed VPS adds a person or a panel on top. The provider patches the OS, watches uptime and often handles the web server and certificates. The price roughly doubles and most of the routine work disappears. The dedicated cloud hosting post covers the isolation end of this spectrum.

A managed platform removes the server from the conversation. You push a repository or pick a WordPress plan; the platform builds, runs, patches, backs up, issues certificates and scales. There is no root and no SSH, which is the trade. For a web app, an API, a database or a site, this is the column most VPS shoppers actually want and do not see in the table, because the table is a table of servers.

The honest test: if you are comparing VPS plans in order to run a web app and you would rather not maintain a server, the comparison you want is between a managed platform’s plans, not between VPS providers.

How a managed plan compares on the same axes

Put a managed platform on the same five columns and the difference is what is included, not the numbers. On RunxBuild, a web service on the $6 Basic plan is 0.5 vCPU and 624MB; the $13 BasicMini plan is 1 vCPU and 1GB; the $20 BasicPlus is 1 vCPU and 2GB; the $65 BasicMax is 2 vCPU and 4GB. Those look like VPS rows. The rows include the build from your repository, the process manager, TLS on a custom domain, runtime logs, rollback to the previous deploy, and autoscaling between a floor plan and a ceiling plan when CPU crosses a threshold, which a VPS cannot do without you rebuilding it.

Databases are on the same ladder as separate managed instances with backups and private networking, so the database is not on the same box competing for memory. A WordPress site has its own ladder starting at $3 a month for the Starter plan.

What you do not get is root. If your comparison includes a custom kernel, a VPN endpoint, a game server or a mail server, a VPS is the right product and this section does not apply. If it includes a web app, an API, a database and a site, price the managed plan next to the VPS plan plus backups plus your hours, and the table looks different.

How this fits the rest of the stack

Compare VPS plans with a workload in hand, read vCPU as shared or dedicated before reading the price, check the bandwidth overage rate and the backup fee, and spend one hourly-billed hour benchmarking the shortlist. Then ask whether the comparison is the right one. For a web app or a site, the RunxBuild hosting calculator puts a managed plan’s line items beside each other, the service, the database, the storage and the bandwidth, so the number you compare against the VPS table already includes the parts the table leaves out.

Useful related references:

FAQ

What is the most important spec when comparing VPS plans?

For most web workloads, whether the vCPU is shared or dedicated, followed by the disk type. A dedicated core and an NVMe disk on a cheaper plan will outperform more shared cores on a SATA SSD for a database or an API. RAM is the easiest number to read and the one people over-weight; size it from the workload first, then compare the other columns.

How do I tell if a VPS is oversold?

Rent it for an hour and run vmstat. The st column is steal time, the share of CPU the hypervisor gave to other tenants while your process wanted it. A steady figure above a few percent means the host is oversold. Run a disk benchmark twice an hour apart; a large swing means shared storage under contention.

Is a managed VPS worth the extra cost?

If you would otherwise spend two or more hours a month on patching, monitoring and troubleshooting, yes, because those hours cost more than the difference. If the server is the point, you enjoy the work, or you need root for something unusual, an unmanaged VPS is the better value. For a plain web app, consider whether a managed platform is the better comparison than either.

What is the difference between a VPS and a managed platform?

A VPS is a server; you install and run everything above the operating system. A managed platform runs your application from a repository and handles the process, TLS, logs, backups and scaling, with no root access. A VPS is cheaper per month and costs hours; a platform is more per month and costs none. The right one depends on whether the server itself is something you want.

Does location matter when comparing VPS plans?

More than most specs. Round-trip latency to your users is set by distance, and fifty milliseconds is a visible difference on every page load. If your visitors are mostly in one region, a slightly slower plan in that region beats a faster one across an ocean. A CDN in front helps for static assets but not for dynamic requests.