Part of the Technology & IT suite · 45 calculators

vCPU Overcommit Calculator

How many virtual machines a host can carry — by CPU at a chosen overcommit ratio and by memory with a reserve for the hypervisor — and which of the two is the limit.

Virtual CPUs are shared: because most VMs idle most of the time, hypervisors let you allocate more vCPUs than there are physical threads.

Results update as you type
Results
VMs the host supports
62
By CPU
By memory
Limiting resource
Effective vCPU ratio at that count
Reviewed September 2026. Computing arithmetic: bytes, bits, seconds and probabilities do not change by country. Ofcom requires UK providers to quote a minimum guaranteed download speed.
No account required · Google Analytics off unless allowedCalculator arithmetic runs in your browserResults update as you type
All calculations run 100% in your browser. The calculator code does not submit your figures to GlobalCalc to obtain a result.
About vcpu overcommit

How the vcpu overcommit calculator works

Virtual CPUs are shared: because most VMs idle most of the time, hypervisors let you allocate more vCPUs than there are physical threads. Ratios of 3:1 to 5:1 are common for general servers, 1:1 for latency-sensitive or licensed-per-core workloads, 8:1 or more for desktops. Memory is not shared the same way, so RAM after the host reserve divided by memory per VM is usually the harder limit.

Formula: VMs by CPU = cores × threads × ratio ÷ vCPUs per VM; VMs by memory = (RAM − reserve) ÷ RAM per VM

Worked examples

InputsVMs the host supportsNote
32 cores, SMT, 4:1, 4-vCPU 8 GB VMs, 512 GB6262 — memory-bound
Database hosts at 1:177 — memory-bound
Virtual desktops at 8:1184184 — memory-bound

Frequently asked questions

What overcommit ratio is safe?

It depends on how busy the VMs are: 4:1 is typical for mixed general servers, 2:1 for busier ones, 1:1 for databases and anything licensed per core. Watch CPU ready time — above 5% the ratio is too high.

Should I count hyperthreads?

Most hypervisors schedule on threads, so yes for the count, but a thread is not a core: two threads give about 1.3× one core’s throughput. Many admins set the ratio against physical cores instead and treat SMT as the margin.

Can memory be overcommitted?

Ballooning and page sharing allow it, but memory pressure hurts far more than CPU contention. Plan on 1:1 and keep the host reserve — 5–10% of RAM plus what the hypervisor needs.

What about failover capacity?

In a cluster, leave enough free on the other hosts to absorb one host’s VMs: with four hosts, run each at no more than 75% of this figure.

Where these figures come from

Last checked: September 2026. Units follow the SI decimal convention (IEC 80000-13 defines the binary alternatives); network and security figures cite the defining standard.