vCPU and RAM pricing: how cloud and VPS compute is billed, and how to right-size it

Compute is priced by the size of the virtual machine (how many vCPUs and how much RAM), the family it belongs to (general purpose, compute-optimised, memory-optimised, burstable, GPU), the region, and how long it runs. Public clouds bill per second or per hour while the machine is running; a VPS bills a fixed monthly price whether you use it or not. The quickest way to cut a compute bill is usually to match the size to real usage and to stop machines that do not need to run around the clock.

This guide sits in the pricing overview. Amounts are illustrative; use the provider's calculator for the Canadian region you will use.

What a vCPU actually is

A vCPU is a share of a physical processor. On many platforms one vCPU is one hardware thread, so two vCPUs may correspond to one physical core with two threads. That matters in two places:

• Performance: a vCPU on one provider or instance generation is not necessarily equal to a vCPU on another. Newer generations often do more work per vCPU at a similar price.
• Licensing: some commercial software is licensed per core rather than per vCPU, and the conversion rules differ. See software licences in the cloud.

RAM is sold in fixed ratios per family. General-purpose families often offer about 4 GB of RAM per vCPU; memory-optimised families more; compute-optimised fewer. If your application needs more memory, moving to a memory-optimised family can be cheaper than buying extra vCPUs you do not use.

Instance families and when they fit

Per-second, per-hour and monthly billing

Public cloud compute is usually billed while the VM is running, even if it is idle. A stopped VM usually stops compute charges, but its disks and reserved IP addresses keep costing money. A VPS is billed for the month regardless.

Illustrative example of the monthly maths. Clouds commonly quote hourly rates and use about 730 hours as an average month (8,760 hours ÷ 12).

• A VM at an illustrative $0.12 an hour, running all month: 730 × $0.12 = $87.60.
• The same VM running only during business hours, 11 hours a day on about 22 working days: 242 hours × $0.12 = $29.04.
• Saving from scheduling: $87.60 − $29.04 = $58.56 a month, about 67%.

Scheduling works for development, test, training and back-office tools. It does not work for a public website or anything customers use at night.

Right-sizing in four steps

1. Measure for two to four weeks. Record average and peak CPU and memory use. Most providers' monitoring tools show this.
2. Look at the peak, not only the average. A server averaging 10% CPU with daily peaks at 85% may already be the right size.
3. Step down one size at a time and watch response times. Sizes typically halve or double, so one step can cut the compute line by about half.
4. Reconsider the family. If memory is the limit and CPU idles, a memory-optimised shape may cost less than a larger general-purpose one.

Illustrative right-sizing. A business runs four VMs at an illustrative $140 a month each ($560). Measurement shows two of them peak below 30% CPU and 40% memory. Moving those two one size down to $70 each saves 2 × $70 = $140 a month, or $1,680 a year, before tax. Commit to a reserved price only after this step; see on-demand vs reserved.

Region and generation affect the rate

Providers price each region separately. A Canadian region can cost more or less than a US region for the same machine, and newer instance generations are often cheaper per unit of work. When you compare, use the same region, family and generation, and write the date of the price next to it. Data location obligations may decide the region before price does; see data residency and Canadian regions.

What compute prices usually do not include

• The disk attached to the VM (priced separately as block storage; see storage tiers, backups and snapshots).
• Public IPv4 addresses, increasingly billed per address per hour.
• Data transfer out (see egress costs).
• Commercial licences, unless you choose a "licence included" image.
• Support plans.

Checklist before you pick a size

• Peak CPU and memory measured, not guessed
• Instance family matches the bottleneck (CPU, memory or burst)
• Schedule set for non-production machines
• Same region and generation used when comparing providers
• Disk, IP, egress and licence lines added before comparing totals

FAQ

Is a bigger VPS always faster?

Not necessarily. If the application waits on a slow disk, a database query or the network, more vCPUs will not help. Measure where the time goes before upgrading.

Do stopped cloud servers cost anything?

Usually the compute charge stops, but attached disks, snapshots and reserved public IP addresses continue to be billed. Delete what you no longer need rather than leaving it stopped for months.

What is the risk with burstable instances?

They are cheap while usage stays below the baseline. Under sustained load they either slow down when credits run out or, in modes that allow unlimited bursting, add charges. They suit spiky, light workloads, not steady processing.

Sources

• Canadian Centre for Cyber Security — security considerations when choosing and configuring cloud services
• Canada Revenue Agency — tax on cloud services you buy
• Bank of Canada — exchange rates for converting USD hourly prices