What does this calculator do?
Memory is usually the first resource to run out on a virtualisation or container host, and unlike CPU it cannot be safely oversubscribed. This calculator applies the two deductions that matter in order: the RAM reserved for the operating system or hypervisor, and a target utilisation ceiling that keeps the host from running at the edge of its memory. What remains is divided by the memory each workload needs, rounded down, because a workload that does not fit entirely does not run. The leftover headroom is reported so you can see how much of the ceiling is unused. Everything is computed in your browser from the values you type. Nothing is looked up on the network — there is no ping, no DNS query, no WHOIS and no geolocation.
Formula
Subtract the reserve, apply the ceiling, then divide and round down.
usable = installedRam − reservedRam planned = usable × (targetUtilisation ÷ 100) workloads = floor(planned ÷ perWorkloadRam) headroom = planned − workloads × perWorkloadRam
How to use it
Enter the installed RAM and the amount reserved for the hypervisor, the operating system and its agents. Set a target utilisation that leaves room for spikes and for failover — 70% to 85% is a common range. Enter the memory a single workload needs, including its own overhead. The result is deliberately rounded down; the headroom figure shows the memory left inside the ceiling, and everything above the ceiling remains as an additional safety margin.
Worked example
A 512 GB server with 32 GB reserved leaves 480 GB usable. At an 80% target that is 384 GB available for workloads. With 16 GB per workload, 384 ÷ 16 = 24 workloads exactly, leaving 0 GB of headroom. If each workload needed 20 GB instead, 384 ÷ 20 = 19.2, so 19 workloads fit and 4 GB of headroom remains.
Frequently asked questions
Why not plan for 100% utilisation?
Because a host at 100% memory has no room for a spike, a restart or a migration, and the moment it needs to swap, performance falls off a cliff. The ceiling is what keeps the host recoverable.
Why is the result rounded down?
A workload either fits in memory or it does not. 19.2 workloads means 19 run inside the plan; the twentieth would push the host past its ceiling.
Does this include failover capacity?
Only indirectly, through the target utilisation. If a cluster must absorb the workloads of a failed host, size the cluster so that one host’s worth of memory stays free on top of what this calculator reports.
Sources
- ISO/IEC 30134-2 — Power Usage Effectiveness (PUE): the standard definition of the data-centre facility-to-IT energy ratio applied to the power results.
- BIPM — The International System of Units (SI Brochure): the watt and the kilowatt-hour, used for the energy and average-load figures.
- Nominal capacity model used here: vCPU capacity = physical cores × oversubscription ratio, and usable RAM = installed RAM − reserved RAM. NUMA topology, hypervisor scheduling, memory ballooning, workload bursts and licence limits are not modelled.
- Reviewed: 2026-08-10
Related calculators
Last updated: 2026-08-10 · v1.0.0 · Report an error