hosting
self-hosting
n8n
vps

Hostinger KVM 2 vs KVM 4 for n8n: which VPS do you actually need?

By Sanjay GuptaUpdated 8 Oct 20267 min read
Rows of server racks in a data centre
Photo: Cern datacenter.jpg by Hugovanmeijeren, CC BY-SA 3.0, via Wikimedia Commons

Disclosure: Some links to Hostinger on this page are affiliate links. If you buy through them we may earn a commission, at no extra cost to you. It does not change which plan we recommend.

Nearly every week someone asks us the same thing: they want to self-host n8n on a VPS, they have seen that KVM 2 costs ₹799 a month and KVM 4 costs ₹1,099, and they want to know whether the extra ₹300 buys them anything they will actually use.

The honest answer for most people is no — and then there is a specific set of workloads where KVM 2 will fall over inside a fortnight and you will wish you had started one plan up. This article is about telling those two situations apart before you pay for two years.

What n8n actually does to a server

n8n is a Node.js application. On a quiet box it sits there using a few hundred megabytes of RAM and almost no CPU. The cost comes in bursts: every time a workflow fires, it spins up an execution, holds the data moving between nodes in memory, writes the execution record to a database, and releases it.

That shape matters because it changes what you should be buying. A web server is mostly limited by CPU and concurrent requests. A self-hosted n8n box is almost always limited by RAM, and specifically by RAM spikes — one workflow pulling a 40 MB file, or looping over 5,000 rows, can briefly ask for more memory than the whole machine has.

Hostinger's own tutorial says this plainly: memory is more critical than CPU for n8n, and you should look at upgrading once memory usage sits above 75–80%. That has matched what we see in practice.

The specs, side by side

Here is what the four KVM VPS plans give you. Prices are the Indian promotional rates on a 24-month term, with the renewal rate next to them — because the renewal is the number that decides whether this was a good deal.

PlanvCPURAMNVMeBandwidthPromo / renewal
KVM 11 core4 GB50 GB4 TB₹599 / ₹999 per month
KVM 22 cores8 GB100 GB8 TB₹799 / ₹1,199 per month
KVM 44 cores16 GB200 GB16 TB₹1,099 / ₹2,399 per month
KVM 88 cores32 GB400 GB32 TB₹2,199 / ₹4,399 per month

Two things are worth noticing before anything else.

First, the gap between KVM 2 and KVM 4 is a doubling of everything for about 38% more money during the promotional term. That is unusually good value on the promo. Second, at renewal the same upgrade costs exactly double — ₹1,199 against ₹2,399. So the decision is not "is KVM 4 worth ₹300 more", it is "is KVM 4 worth ₹1,200 more every month from year three".

What n8n's own requirements say

n8n's documentation and Hostinger's VPS guidance converge on roughly the same numbers. The floor for running n8n at all is around 1 vCPU, 2 GB of RAM and 20 GB of storage — fine for testing and personal projects with a handful of simple workflows. For a production setup that a business depends on, the recommendation moves to 2–4 vCPU, 4–8 GB RAM and 40–80 GB of NVMe.

Read that range against the table above and the answer falls out: KVM 2 (2 vCPU, 8 GB) sits at the top of the recommended production band. It is not the minimum. It is already comfortable.

When KVM 2 is genuinely enough

KVM 2 will handle all of this without complaint:

  • Twenty to fifty active workflows, firing on schedules or webhooks.
  • Lead routing, CRM syncs, WhatsApp and email follow-up, Google Sheets updates — the everyday glue work most people build n8n for.
  • A few thousand executions a month, each moving JSON rather than big files.
  • OpenAI, Claude or Gemini API calls. These are network-bound, not CPU-bound — your server is waiting on someone else's GPU, not working.
  • n8n plus a small Postgres instance on the same box, plus Nginx and SSL.
  • A second light application alongside it, such as a staging site or a metrics dashboard.

If your automation looks like that — and for most Indian agencies, D2C brands and SaaS teams it does — the ₹300 you would spend upgrading is better spent on API credits.

The cheapest VPS upgrade is the one you don't buy because you configured the box properly.

When you genuinely need KVM 4

There are five workloads that break KVM 2, and all of them are memory stories rather than CPU stories.

1. Browser automation inside n8n

If your workflows run Puppeteer, Playwright, or any node that drives a headless Chrome, stop reading and buy KVM 4. A single headless Chrome instance can take 500 MB to 1 GB on its own, and scraping workflows rarely run one at a time. This is the single most common reason we see an n8n box die.

2. Large binary files

PDFs, video, image batches, big CSV exports. By default n8n holds binary data in memory as it passes between nodes. A workflow processing 200 invoices at 3 MB each is asking for well over a gigabyte in one execution — before you count the Node.js overhead around it.

3. Queue mode with multiple workers

Once you move to n8n's queue mode you are running a main process, Redis, and one or more worker processes, each of which is a full Node.js runtime. Two workers plus Redis plus Postgres is a four-service box. 8 GB gets tight; 16 GB does not.

4. High-concurrency webhooks

If you are receiving webhooks from a busy platform — payment gateways, a WhatsApp Business API provider, a marketplace order feed — the problem is not daily volume, it is the spike. Three hundred orders arriving in ten minutes during a flash sale will queue up executions faster than two cores can clear them.

5. n8n is not alone on the box

Running n8n alongside a real WooCommerce store, a Metabase instance, or a few client WordPress sites is a completely different sizing problem. At that point you are buying a server, not an n8n host, and 16 GB is the sane floor.

The settings that keep you on KVM 2

Before upgrading, change these four things. We have moved more than one workload back down a plan this way.

  1. Prune execution data. By default n8n keeps every execution record forever, and the database quietly grows until queries slow down. Set EXECUTIONS_DATA_PRUNE=true and a max age of 7–14 days unless you have an audit reason to keep more.
  2. Move binary data out of memory. Setting the binary data mode to filesystem means large files are written to disk instead of held in RAM between nodes. You have 100 GB of NVMe; use it.
  3. Cap concurrency. n8n lets you limit how many executions run at once. Setting a sane ceiling turns a crash into a short queue, which is almost always the better failure.
  4. Add swap. A 2–4 GB swap file will not make your server fast, but it will stop a one-off memory spike killing the Node process at 2am. This takes two minutes on Ubuntu and nobody does it.

Do these four and watch memory usage for a fortnight. If it still sits above 80%, the upgrade is real and you should take it.

What the two years actually cost

On the 24-month term, KVM 2 comes to roughly ₹19,176 and KVM 4 to roughly ₹26,376 — a difference of about ₹7,200 across two years, or ₹300 a month. Set against one developer-hour, that is nothing. Set against renewal, it is a different story: from month 25, the same two plans cost ₹1,199 and ₹2,399 a month.

So the practical advice is: buy for the next 24 months honestly, not aspirationally. Hostinger lets you scale a VPS up without reinstalling — you pay the difference and reboot. There is no reinstall, no migration, no re-pointing DNS. Downgrading is the awkward direction, not upgrading. That asymmetry is the whole argument for starting on KVM 2.

A ten-minute test that settles it

If you already run n8n anywhere — even locally — do this before you buy:

  1. Run your three heaviest workflows back to back.
  2. Watch memory with htop or free -m while they run.
  3. Note the peak, not the average.

If your peak is under about 3 GB, KVM 2 has roughly 2.5× headroom and you are fine. If the peak is over 5 GB, or you cannot see a peak because the process was killed, start at KVM 4. Guessing from workflow count is unreliable; one workflow that touches files beats forty that touch JSON.

Our recommendation

Start on KVM 2 unless you run browser automation, move large files, or need queue mode. Those three cases take you to KVM 4 immediately. Everyone else should take the cheaper plan, apply the four settings above, and keep the difference.

If you are still deciding between self-hosting at all and paying for the managed version, the two are compared properly in Hostinger vs n8n Cloud — including the maintenance hours that never appear on a pricing page.

And while you are debugging workflows, a few browser helpers save real time: a JSON viewer for reading a node's output, a base64 decoder for webhook payloads, and a password manager that generates and stores the admin credentials you are about to set on that fresh server. You will find all of them in the developer extensions category.

Frequently asked questions

Is KVM 1 enough for n8n?

It can be. KVM 1 gives you 1 vCPU and 4 GB of RAM, which comfortably exceeds n8n's documented minimum of 1 vCPU and 2 GB. It is fine for testing, personal projects and a handful of simple scheduled workflows. For anything a business depends on, KVM 2's second core and 8 GB of RAM give you the headroom to survive a workflow that misbehaves.

Can I upgrade from KVM 2 to KVM 4 later?

Yes, and this is the main reason to start small. Hostinger lets you scale a VPS up by paying the difference and rebooting — there is no reinstall, no migration and no DNS change. Downgrading is the awkward direction, so the risk of starting on KVM 2 is far lower than the risk of over-buying.

How many n8n workflows can KVM 2 handle?

Workflow count is the wrong unit. A box can run a hundred lightweight JSON workflows and fall over on three that process PDFs. As a rough guide, KVM 2 handles twenty to fifty active workflows and a few thousand executions a month without complaint, provided none of them run headless browsers or hold large files in memory.

Does running AI nodes need more RAM?

Not if you are calling an API. OpenAI, Claude and Gemini nodes send a request and wait — the heavy computation happens on someone else's hardware, so your server is idle. RAM becomes the constraint only if you run a model locally, which is a different sizing exercise entirely.

Related on Single Tab Browser

Go deeper on the extensions mentioned in this post.

More from the blog

Posts on nearby topics, so nothing on the blog sits without a way in.

Browse the directory