n8n VPS requirements · measured, not guessed
n8n VPS requirements:
test your peak before you buy.
There is no honest universal RAM number for every n8n workflow. Enter a candidate server and your own representative test. This planner checks headroom, concurrency, latency, recovery, monitoring, and who operates the box.
- No signup
- No server connection
- Browser-only inputs
- No affiliate tracking
PURCHASE GATE
Retest the real peak
RAM guessNot accepted
Required inputOwn peak measurement
Production gateLoad + restore + owner
● A BLOG NUMBER CANNOT PASS YOUR WORKLOAD
The useful answer is a test result
Minimum specs start a trial. They do not prove production fit.
n8n’s own documentation says performance varies with workflow type, available resources, and scaling configuration. Binary data, Code nodes, manual runs, and simultaneous workflows can all change memory demand. A buying decision must cover your actual peak.
What we actually measured
One bounded n8n 2.37.7 fixture
- 2,500 synthetic items, five successful repetitions
- 8.428 s median end-to-end CLI time
- 585.3 MiB maximum observed process-tree working set
- Controlled failure, same-workflow recovery, and SQLite restore passed
What that number cannot establish
Your VPS requirement
- It was not a Linux VPS or a hosting-provider benchmark
- It did not use your nodes, payloads, files, or concurrency
- It does not include your database, proxy, monitoring, or backup stack
- It is a reference measurement, not a 1 GB, 2 GB, or 4 GB recommendation
Interactive purchase gate
Test a candidate VPS without pretending a guess is evidence.
Use the public lab only to understand the method. A candidate can pass only when the measurement comes from your own representative workload and every production gate is checked.
A six-step sizing runbook
Make the purchase follow the evidence.
Run this before a long commitment. Repeat it after materially changing nodes, payloads, concurrency, execution retention, database, or n8n version.
- 01
Define the peak and service target
Write down the largest payload, planned simultaneous executions, acceptable p95 latency, and allowed error rate. Do not size from monthly executions alone.
- 02
Build a safe representative fixture
Use synthetic or redacted records that preserve payload size, branching, binary handling, Code nodes, and external-call timing without exposing credentials or customer data.
- 03
Limit the complete candidate stack
Include n8n, database, reverse proxy, queue or worker components, monitoring, and backup processes inside the resource boundary you intend to buy.
- 04
Run the production mode repeatedly
Exercise the planned trigger path at or above peak concurrency. Capture process-tree memory, swap, CPU, p95 latency, errors, restarts, and queue delay for at least five clean runs.
- 05
Break and restore it
Create a controlled failure, verify alert delivery, correct or roll back the workflow, then restore a clean backup into an isolated target and compare the restored material.
- 06
Buy only the passing operating model
If the resource gate fails, test a larger candidate or reduce the workload. If recovery or ownership fails, choose managed hosting or n8n Cloud instead of disguising an operations gap as cheap VPS savings.
If the candidate passes
Compare the exact VPS contract.
Verify dedicated RAM, scaling path, backups, snapshots, monitoring, renewal price, region, support boundary, and whether root access matches your plan.
Review current VPS specifications Direct provider link. No affiliate parameter or tracking is active.If operations fail
Do not buy yourself an unpaid on-call job.
Use the full decision lab to compare n8n Cloud, contract-defined managed hosting, and self-managed VPS with operator labor separated from cash price.
Open the hosting decision lab Recommendations remain provisional until load and recovery tests pass.Primary sources, checked 2 Sep 2026
What this gate is built on.
Performance factors
n8n documents that workflow type, available resources, and scaling configuration affect performance, and points users to their own benchmarking.
n8n performance documentationMemory factors
n8n identifies binary size, node count, Code nodes, manual runs, and simultaneous workflows as factors that can increase memory use.
n8n memory documentationConcurrency
Self-hosted regular mode does not limit production concurrency by default; n8n documents a production concurrency control that queues excess work.
n8n concurrency documentationSupport boundary
A VPS provider and n8n do not automatically own the same layer. Confirm who handles infrastructure, OS, n8n updates, backups, and workflow incidents.
n8n VPS support scopeSource links are direct and untracked. Current product specifications, pricing, and documentation can change; verify them again before purchase.
Questions that change the server decision
n8n VPS requirements FAQ
Is 2 GB or 4 GB RAM enough for n8n?
It may be enough for some workloads and insufficient for others. The responsible answer requires your full stack, largest representative payload, production trigger mode, and peak concurrency. Use a candidate size as a test boundary, not a guarantee.
Why can’t I multiply the 585.3 MiB lab result?
The observed figure belongs to one Windows process tree, one n8n version, one synthetic workflow, and one concurrency profile. Memory does not scale reliably from that fixture to different nodes, files, databases, or parallel executions.
Does monthly execution count determine RAM?
No. Monthly volume affects plan economics, but memory pressure is driven more directly by what a run processes and how many runs overlap. A low monthly total can still contain a large burst.
When should I choose n8n Cloud instead?
Choose the managed route when published limits fit and your team should not own server patching, monitoring, backup restoration, and incidents. Choose self-managed VPS only when control is required and an operator can prove and maintain the stack.
Does a snapshot count as a tested backup?
Not until it restores into a clean target and the required workflows, credentials path, database material, and run behavior are verified. Recovery evidence is an executed result, not the existence of a file.
Will this tool send my server data anywhere?
No. It is static JavaScript with no account, contact form, analytics input endpoint, or submission target. Your entries remain in the current browser page.