> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rafftechnologies.com/llms.txt
> Use this file to discover all available pages before exploring further.

# CPU and bursting

> The sustained level Standard plans can always hold, how bursting above it works, and what happens after long periods of heavy load

<sub>Updated September 14, 2026</sub>

Standard VMs are **burstable**. Your VM can use all of its vCPUs whenever it needs them, and it has a **sustained level** it can hold indefinitely. Most VMs never reach the limit. Bursting is what makes Standard both fast and cheap.

This page applies to **Standard (General Purpose)** plans. [Premium and Windows Server plans](/products/build/virtual-machines/concepts/plans-and-sizing) use guaranteed vCPU and work differently.

## How it works

Your plan includes a sustained level of **40% of your vCPUs**. On a 4 vCPU VM that is 1.6 cores, available every second of every day.

Use less than that and you build credit. Use more and you spend it. Bursting above your sustained level is included. It is not an extra charge, and there is nothing to enable.

Credit builds up to **24 hours' worth**, so a VM that is quiet overnight starts the morning with a full balance.

### When credit runs out

After a long stretch of heavy use the balance reaches zero. Your VM then alternates: it runs at a reduced level (**20% of your vCPUs**) for about **two hours** while credit rebuilds, then returns to full speed automatically.

Nothing is suspended, nothing needs restarting, and no action is required from you. Averaged over time this works out to your sustained level.

## How much burst you get

Two cases, and which one you are in depends only on your average CPU use over time.

**Below 40% average, nothing ever happens.** The reserve only fills. Your VM runs at full speed permanently and never sees a limit. Most VMs live here.

**Above 40% average, your VM averages 40%.** You still get the extra, but delivered in bursts rather than continuously. Under constant heavy load a VM alternates: roughly two hours at its sustained level, then a stretch at full speed, then the same again.

A VM that has been quiet starts with a full reserve, so it can run flat out for about **16 hours** before any of this applies. That is a head start, not something you get back each day.

Short spikes never come close. A build that pins every core for twenty minutes, a nightly backup, a traffic burst: none of them touch the reserve in any meaningful way.

## Where to see it

The **CPU** card on your VM's Information tab shows the current state:

| State           | Meaning                                             |
| --------------- | --------------------------------------------------- |
| **Full speed**  | All cores available. Nothing to think about         |
| **Bursting**    | Currently using more than your sustained level      |
| **At baseline** | Credit is rebuilding. Full speed returns on its own |

The card also shows the last 24 hours, so a pattern is easy to spot.

### What it looks like inside your VM

If you run `top` while your VM is at its sustained level, you will see a high
**steal time** (`st`), with load average unchanged:

```
load average: 4.16, 3.50, 1.91
%Cpu(s):  7.8 us,  9.5 sy,  0.0 id,  82.8 st
```

Steal time usually means a busy host, so this is easy to misread. Here it means
something simpler: your processes are asking for CPU and are being held to your
sustained level while the reserve refills. Load average stays high because the
work is still queued. Nothing is wrong with the host, and nothing is wrong with
your VM.

## If you need more

A VM that sits at **At baseline** regularly is telling you something. Two options:

* **Resize to more vCPUs.** The sustained level is a percentage, so it scales with the plan. An 8 vCPU Standard VM holds 3.2 cores indefinitely. See [Resize a VM](/products/build/virtual-machines/quickstart-guides/resize).
* **Move to Premium.** [CPU Optimized plans](/products/build/virtual-machines/concepts/plans-and-sizing) use guaranteed vCPU, which suits workloads that need consistent CPU every millisecond: databases, real-time services, latency-sensitive APIs.

## Related

* [Plans & sizing](/products/build/virtual-machines/concepts/plans-and-sizing): choosing between Standard, Premium and Windows Server
* [Monitoring & metrics](/products/build/virtual-machines/concepts/monitoring-and-metrics): CPU graphs and history
* [Pricing](/products/build/virtual-machines/details/pricing): full price list