Prices, bill drivers and 2026 rules

vCPU and RAM pricing: how cloud server and VPS sizes are charged

How vCPU, RAM, instance families, burstable servers, operating system licences and running hours set the compute price of a cloud server or VPS, with worked sizing examples.

Updated · 5 min read

vCPU and RAM pricing: how cloud server and VPS sizes are charged

Compute is priced by size × hours × family. A cloud server's hourly price rises with the number of virtual CPUs (vCPU) and the amount of RAM, changes with the instance family (general purpose, compute-optimised, memory-optimised, burstable), includes a licence charge for Windows and some databases, and runs for every hour the server exists in a running state. A VPS bundles a fixed size into a monthly price. The quickest saving is usually not a cheaper provider but a right-sized server that runs only the hours it needs: halving the size or the running hours roughly halves the compute line.

This guide is part of the pricing overview. Commitment discounts are in on-demand vs reserved vs spot, and the type-level comparison is in shared vs VPS vs dedicated vs cloud. Figures are illustrative, ex GST.

What a vCPU and RAM actually buy

A vCPU is usually one thread of a physical processor core. Two vCPUs are often one physical core with two threads, though providers describe this differently, and burstable types give you only a fraction of a vCPU's sustained capacity. RAM is the memory the server can use; databases and caching need more of it than web servers. Providers group sizes into families with set ratios of RAM to vCPU:

Family (generic name)Typical RAM per vCPUFitsPrice tendency
Burstable0.5–4 GB, with CPU creditsLow-traffic sites, dev and test, small toolsLowest, until credits run out
General purposeAbout 4 GBWeb and application serversBaseline
Compute-optimisedAbout 2 GBCPU-heavy batch work, encodingLower per vCPU, less RAM
Memory-optimisedAbout 8 GB or moreDatabases, caches, analyticsHigher per vCPU
GPU or acceleratedVariesMachine learning, renderingMuch higher

Ratios are generic patterns; check the provider's own instance documentation. A newer processor generation is often cheaper for the same performance, so an old instance type can cost more than its replacement.

Running hours: the multiplier people forget

Cloud compute is billed per second or per hour while the server is running. A month is roughly 730 hours (8,760 hours ÷ 12).

Illustrative example: a general-purpose server at $0.20 an hour.

Running patternHours a monthMonthly compute
Always on730730 × $0.20 = $146.00
Business hours, 7 am–7 pm weekdays12 × 22 = 264264 × $0.20 = $52.80
Test server, 4 hours a day on weekdays4 × 22 = 8888 × $0.20 = $17.60

Switching a test or staff-only server off out of hours cuts its compute line by about 64% here (1 − 264 ÷ 730). Stopped servers usually still pay for their disks and any reserved public IP address. Production websites and databases generally need to stay on.

Operating system and software licences

Linux images are commonly priced without a licence charge. Windows Server images usually include a licence cost in the hourly price, and SQL Server or other commercial databases add more. Some providers let you bring your own licence, but licence rules for running on shared cloud hardware are set by the software vendor and can be complex. A rough method:

  1. Price the same size with Linux and with Windows in the same region.
  2. The difference is the licence component for that size.
  3. Multiply by running hours and server count.

If the difference is large, ask whether the application genuinely needs Windows, or whether a managed database service is cheaper than a licensed database on a virtual machine. See support plans and software licences.

Sizing without guessing

  1. Measure the current server. Average and peak CPU use, memory use and disk input/output over at least a month, including month-end or seasonal peaks.
  2. Size to the peak you must serve, not the theoretical maximum. If CPU averages 15% and peaks at 45% on a 4-vCPU server, a 2-vCPU server would peak near 90%, which is tight. Try a larger burstable size or keep 4 vCPU and consider a memory change instead.
  3. Separate roles. A database usually needs more RAM; a web server more instances. Splitting them can let you right-size each.
  4. Review after a month. Cloud resizing is usually a reboot away. Most providers offer right-sizing recommendations in their cost tools.

Illustrative right-sizing: a 4-vCPU, 16 GB server at $0.40 an hour always on costs 730 × $0.40 = $292 a month. Measurement shows memory at 5 GB and CPU peaks of 30%. A 2-vCPU, 8 GB server at $0.20 an hour costs $146 a month: a saving of $146 a month or $1,752 a year, before any commitment discount.

VPS plans: same logic, bundled

A VPS plan bundles vCPU, RAM, disk and a data allowance into one monthly price. Compare plans on:

  • vCPU type (dedicated or shared threads) and RAM;
  • disk size and type (SSD or NVMe, and any input/output limits);
  • included data transfer and the price per GB beyond it;
  • whether backups, snapshots and a public IPv4 address are included;
  • whether the price is for a promotional first term.

Frequently asked questions

Is more vCPU always faster?

No. Many small business applications are limited by memory, disk speed or database queries rather than CPU. Measure before you upsize.

What is a burstable instance?

A low-cost size that earns CPU credits while idle and spends them when busy. If it runs busy all the time, it either slows down or charges for extra credits, depending on the provider and settings.

Do stopped servers cost anything?

Usually their disks, snapshots and any reserved IP addresses keep charging. Delete what you do not need, after confirming backups.

Should I pay more for a newer processor generation?

Often the newer generation costs the same or less per hour and performs better, so it can lower the cost per unit of work. Check the provider's current instance pages.

Sources

Related guides