VM Host Sizing Calculator
Estimate VMs per host from physical cores, vCPU oversubscription, RAM, reserves and average VM size. See CPU and memory limits separately, with N+ failover headroom.
Result
| Physical cores/host | 32 |
|---|---|
| Usable vCPU capacity/host | 108.8 vCPU |
| Usable RAM/host | 480.0 GiB |
| VMs by CPU | 27 |
| VMs by RAM | 60 |
| Recommended VMs/host | 27 |
| Hosts for workload | 4 (ceil(100 / 27)) |
| Total hosts with N+1 reserve | 5 hosts (4 + 1 reserved) |
| VM capacity if 1 host fails | 108 VMs |
Note: CPU ratio is total allocated vCPU divided by physical cores. RAM is not overcommitted here. Actual density depends on workload peaks, NUMA, storage and network limits, hypervisor overhead, licensing, and measured contention. Total hosts = ceil(total VMs / VMs per host) + reserved hosts, so the workload still fits if the reserved number of hosts fail. Assumes identical hosts.
Learn more
How it works
A host runs out of either CPU or memory first. The calculator works out both and uses the smaller:
- CPU: physical cores x (1 - reserve) x vCPU:pCPU ratio = vCPUs you can hand out. Divide by the vCPUs per VM.
- Memory: host RAM - reserved RAM, divided by RAM per VM. Memory is not overcommitted here.
Hosts for the workload = total VMs / VMs per host, rounded up. Then it adds the hosts you want in reserve (N+1, N+2) so the VMs still fit after a failure.
Worked example
The default is 2 sockets x 16 cores, 15% CPU reserve, a 4:1 vCPU ratio, 512 GiB RAM with 32 GiB reserved, VMs of 4 vCPU and 8 GiB, 100 VMs and N+1:
- CPU: 32 x 0.85 x 4 = 108.8 vCPU, so 27 VMs
- Memory: 480 GiB / 8 = 60 VMs
- Per host: 27 (CPU is the limit)
- Hosts: 100 / 27 = 3.7, so 4, plus 1 spare = 5 hosts
Since CPU runs out long before memory, a lower vCPU ratio or bigger CPUs would change the answer. Adding RAM would not.
What vCPU:pCPU ratio should I use?
General server workloads often run at 3:1 to 5:1. Busy databases and latency-sensitive VMs are often kept close to 1:1. Start low and raise it once you can watch CPU ready time.
Should I count hyperthreads as cores?
This calculator uses physical cores. Hyperthreading adds some throughput but nowhere near a second core, so leave it out of the core count and let the ratio absorb it.
Why reserve a whole host?
If a host fails or goes down for patching, its VMs restart on the others. N+1 means the cluster still holds every VM with one host missing.
Maintained by the RACKCRUNCH team. Last reviewed September 29, 2026. How we calculate | Report a problem