n8n system requirements: how much CPU, RAM and disk you need

By the serverserv team · 7 min read · Updated 22 October 2026

n8n itself is light: an idle instance uses around 100 MB of memory, and n8n's own guidance is 320 MB to 2 GB of RAM for the app. What decides your server size is the work you give it: how many workflows run at once, how much data each one moves, and whether you run a database and AI steps on the same machine. For most small teams, 2 vCPU and 4 GB RAM is the comfortable starting point. This guide gives sizes by workload and the settings that stop n8n running out of memory.

Quick answer: sizes by workload

Workload	vCPU	RAM	Disk
Trying n8n, a few simple workflows	1	2 GB	20 GB
Small business: dozens of workflows, webhooks, some AI steps	2	4 GB	50 GB
Busy team: hundreds of runs an hour, larger files, PostgreSQL on the same server	4	8 GB	100 GB
Heavy: big data sets, many parallel runs, queue mode with workers	6 to 12	12 to 48 GB	200 GB or more

These are practical starting points, in line with sizes the n8n community commonly recommends. Use NVMe or SSD storage: n8n writes every execution to its database, so slow disks slow everything down.

What n8n says

n8n's own prerequisites list memory of 320 MB to 2 GB for the application and an SSD-backed database of 512 MB to 4 GB to start, and note that memory requirements usually matter more than CPU. An idle instance uses about 100 MB. Those numbers are for n8n alone. On a VPS you also run the operating system, a reverse proxy for HTTPS and often PostgreSQL, which is why 2 GB is the realistic minimum for a server and 4 GB is where most people stop worrying.

Why RAM matters more than CPU

Most n8n work is waiting on other services: an API call, a database query, an AI model. That uses little CPU. Memory is different, because n8n holds the data passing through a workflow in memory while it runs. Three things push memory up fast:

• Large items or files: a workflow that downloads PDFs, images or big spreadsheets keeps them in memory unless you change the binary data mode (below).
• Code nodes on big data sets: looping over tens of thousands of items in one Code node can use all the memory available.
• Many runs at the same time: ten webhook calls arriving together means ten executions in memory at once.

SQLite or PostgreSQL?

n8n stores workflows, credentials and execution history in SQLite by default, in a single file under ~/.n8n. That is fine for one person and light use. For a team or anything in production, n8n recommends PostgreSQL, and supports the latest major versions (16, 17 and 18 at the time of writing).

• SQLite: no setup and nothing extra running, but it struggles with many simultaneous runs and cannot be used for queue mode.
• PostgreSQL: handles concurrency well, is easier to back up while running, and is required for queue mode. On the same server it adds roughly 0.5 to 1 GB of RAM to plan for.

Disk space

The n8n app and Docker images take a few GB. What grows is execution history: every run is saved with its data. n8n prunes old executions by default, but if you keep long history, save every successful run or handle files, plan 50 GB or more and watch it. Keep binary files on disk rather than in the database (below) and back up both the database and the .n8n folder, which holds your encryption key.

Stopping out-of-memory crashes

If n8n restarts in the middle of a workflow, or the logs show "JavaScript heap out of memory", work through these in order:

1. Store files on disk: set N8N_DEFAULT_BINARY_DATA_MODE=filesystem so files are written to disk instead of being held in memory.
2. Process in batches: use the Loop Over Items node to handle a few hundred items at a time instead of everything at once.
3. Split big workflows: move heavy steps into a sub-workflow; its memory is released when it finishes.
4. Save less execution data: stop saving data for successful runs of high-volume workflows, and keep pruning on.
5. Add RAM: if the server is regularly above 80% memory, move up a size. It is usually the cheapest fix.

When to move up a size

• Memory use sits above 80% most of the day, or the server starts using swap.
• Webhook replies get slow when several calls arrive together.
• You add PostgreSQL, a local AI model or another app to the same server.
• You need queue mode with separate workers for high volume.

serverserv n8n hosting starts at 2 vCPU, 4 GB RAM and 50 GB NVMe with n8n pre-installed, for $14.99 a month or $10.99 a month on a 3-year term, and goes up to 18 vCPU and 96 GB RAM. You can move up a plan without reinstalling. Setting it up yourself instead? Follow our self-hosting guide and security checklist.

Questions

Sources

• n8n Docs: Prerequisites (memory and CPU)
• n8n Docs: Choose a database
• n8n Docs: Self-hosting
• n8n Community: server sizing discussion