CloudCostCompareBurstable vs fixed-performance VMs: why we never mix them
Some virtual machines can use their full CPU all the time. Others are sold at a lower price on the understanding that they usually use only a fraction of it, and may use more for short bursts. Treating the two as the same product gives wrong comparisons in both directions.
How burstable instances work
A burstable instance has a baseline: the share of each vCPU it is entitled to use continuously. While it runs below the baseline it earns credits. When it needs more, it spends credits to run faster. If it runs out of credits, it is held to the baseline, or on some plans billed extra to keep going.
AWS calls its families T-series, Azure calls them B-series, and Google calls them shared-core machine types. Oracle offers burstable options on its flexible shapes, set with a baseline percentage. The mechanics and the exact baseline differ, so read the vendor’s page for the family you use.
When burstable is the right choice
- Idle most of the time, with occasional peaks: small websites, development servers, internal tools, build agents that wait.
- A low baseline that fits your average load. If your average CPU use stays below the baseline, you never run out of credits.
- You value a low hourly price and can accept variable speed.
When it is a trap
- Sustained load. A job that keeps the CPU busy for hours drains the credits and then runs slowly, or costs more than a fixed-performance instance would have.
- Latency-sensitive services at steady traffic. The slowdown arrives without a crash, which makes it hard to spot.
- Benchmarks run on a fresh instance. A new burstable instance often starts with credits, so a short test looks faster than real life.
How to tell which you need
Watch two numbers for a week or two: average CPU use and the credit balance. If the balance stays high, a burstable size is fine. If it keeps falling toward zero, move to a fixed-performance size. Most clouds expose the credit balance as a metric you can alarm on.
Why this site never compares them
Our rule is to compare only equivalent services. A burstable instance and a fixed-performance one with the same vCPU and memory deliver different amounts of work, so putting them side by side makes the burstable one look far cheaper than it is. The estimator’s compute line uses a fixed-performance family on each cloud, so the totals are comparable. If you use burstable sizes in production, your real bill for those servers will be lower than ours and your performance will vary.
A practical rule
For a service that must perform steadily, choose fixed performance. For a workload that is mostly idle, test a burstable size and watch its credits. In the price catalog you can filter by name or family to find both kinds and compare their prices knowing they are different products. For general sizing advice, see choosing a cloud VM size.