Run Ollama on a Budget CPU VPS — No GPU Required
Contabo's Cloud VPS Plus line is CPU-only and inexpensive — enough to run a 7B-8B quantized model without the GPU-VPS price tag. Here are the exact panel steps, verified directly against Contabo's own documentation.
Written by Tom George · How we research hosts
About This Guide
Data transparency: Order flow, panel navigation, SSH key setup, firewall configuration, and plan pricing in this article are sourced directly from Contabo's own help center and product pages, verified by direct fetch. Two items — the exact SSH-key field screenshot and real CPU throughput numbers — are flagged below as pending a live test-account pass. We earn a commission if you sign up through our links — this never changes our verdict.
Cloud VPS Plus 4 is the entry point this guide builds on: 4 vCPU and 8 GB RAM, CPU-only, at €13.50/mo list (€10.80 for the first 24 months).
It suits a personal setup running a 7B-8B quantized model. For a 70B-class model or multi-user production inference, a GPU VPS is the better fit.
Comparing providers first? Our Best VPS for AI guide compares Contabo, Hostinger, Hetzner, Vultr and DigitalOcean.
Who This Guide Is For (And Who Should Get a GPU VPS Instead)
Ollama runs on CPU. It doesn't need a GPU to work at all — it needs a GPU to work fast at scale. For a single user running a 7B or 8B model for personal projects, a CPU-only VPS is a perfectly reasonable way to get started without paying a GPU-tier premium.
That's the gap this guide fills. Contabo's own blog already has a detailed GPU VPS inference guide aimed at larger models and production workloads — we're not trying to replace that. If your use case matches theirs, read Contabo's GPU VPS guide instead of this one.
If you're closer to "I want to try running a small model on my own server without committing to a GPU bill," keep reading — the rest of this guide is the exact Contabo panel steps to get there.
Step 1: Order a Cloud VPS Plus 4
Contabo markets this product line as "Performance VPS," but the plan you'll actually see and select at checkout is named Cloud VPS Plus N. We use the real SKU name throughout this guide so it matches what you see in your own panel.
| Plan | vCPU / RAM | NVMe / Price |
|---|---|---|
| Cloud VPS Plus 4 | 4 vCPU / 8 GB | 150 GB / €13.50/mo* (€10.80 promo) |
| Cloud VPS Plus 6 | 6 vCPU / 12 GB | 300 GB / €19.00/mo* (€15.20 promo) |
*List price. Contabo shows €10.80 (Plus 4) and €15.20 (Plus 6) for the first 24 months, including applicable taxes. Source: contabo.com/en/vps-performance/, verified 28 September 2026.
Longer plans exist above Plus 6 if you want more headroom for a second model — see Contabo's pricing page directly, as tier pricing is the kind of number that drifts and should be checked at time of purchase.
For this guide, Cloud VPS Plus 4 is the entry point we're building on — 8GB RAM is the realistic floor for running a 7B-8B model without the OS and Ollama's own overhead crowding it out.
The order flow itself
Contabo's checkout is five steps: pick the product and click "Get Started," configure billing period / data center / OS / add-ons, set your server password, enter account details, then pay.
One detail worth flagging before you start: your server password is not stored or emailed to you. Write it down somewhere safe the moment you set it — lose it and the only recovery path is a full reinstall.
After payment, you'll get two emails: your Customer Panel login, and a separate "Your login data!" email with your server's IP address and VNC access details. Provisioning is automatic from there.
If you want to follow along on your own server, order a Cloud VPS Plus 4 now — the rest of this guide picks up right after your server finishes provisioning.
Step 2: First Login and SSH Key Setup
Once your server is provisioned, everything happens in the Customer Panel at new.contabo.com. The path to your server is Servers & Hosting → VPS → [your server] → More (the three-dot menu) → Install/Reinstall.
On the reinstall screen, choose Standard installation, pick your OS, and set a password. This same screen is also where you can preload an SSH public key before the install runs — Contabo's own documentation confirms this field exists on the reinstall screen specifically, not at initial checkout.
If you'd rather add your key manually after first logging in with your password, that works too:
ssh-keygen -t ed25519
ssh-copy-id root@YOUR_SERVER_IPOr, if ssh-copy-id isn't available on your system:
mkdir -p ~/.ssh
chmod 700 ~/.ssh
echo "YOUR_PUBLIC_KEY" >> ~/.ssh/authorized_keys
chmod 600 ~/.ssh/authorized_keysEither path gets you to the same place — key-based login instead of typing your password every time. With that sorted, you're ready to install Ollama itself.
Step 3: Install Ollama
SSH into your new server, then run Ollama's own install script:
curl -fsSL https://ollama.com/install.sh | shThis is the same command Contabo's own blog uses in its Ollama-related posts, and it comes straight from Ollama's own domain — nothing Contabo-specific about it. It installs the Ollama binary, creates a system user, and sets up a systemd service that starts on boot.
Step 4: Pull and Run a Small Model
With Ollama installed, pull a small quantized model — something in the 7B-8B range fits the Cloud VPS Plus 4's 8GB RAM:
ollama pull llama3.2:8b
ollama run llama3.2:8bExpect noticeably slower generation than GPU inference — that’s the trade-off for the lower price. Exact throughput depends on model size, quantization, and what else is running on the server at the time.
We haven’t run a live benchmark on a Cloud VPS Plus 4 yet, so we’re not going to hand you a tokens-per-second number we can’t back up. We’ll update this section with real numbers once we’ve run it live — until then, treat any specific throughput claim you see elsewhere for this spec as an estimate, not a measured result.
If the response feels too slow for your use case, the fix isn't a faster CPU tier on the same product line — it's a GPU VPS, which is a different pricing bracket entirely. See the link in the first section above.
Step 5: Opening the Firewall for Remote API Access
By default, Ollama's API listens on port 11434, and if you're only calling it from the same server (an n8n instance running alongside it, for example), you don't need to open anything — localhost access already works.
If you genuinely need to reach that API from another machine, Contabo's Firewall is a free, panel-managed feature that's off by default and blocks all inbound traffic once you turn it on. Here's the exact path:
The panel click-path
- In the Customer Panel, go to Network Services → Firewalling.
- Click Create firewall, give it a name, and confirm.
- Click into the firewall you just created, open the Inbound Rules tab, and click Add rule to firewall.
- Set Type of Traffic to Custom (11434 isn't one of the predefined options like SSH or HTTPS), Protocol to TCP, and Port(s) to
11434. - Click Add Rule, then go to the Active VPS/VDS tab inside the firewall and add your server. Rules apply immediately once assigned.
The Source(s) field is the one setting in this whole guide that’s a real security decision, not a formality. Ollama’s API has no built-in authentication — anyone who can reach port 11434 can query your model and use your server’s compute for free.
Set Source to your own IP address or a specific CIDR range, never “Any” or “Any IPv4.” If your IP changes frequently, a small CIDR range covering your ISP’s typical block is safer than opening the port to the entire internet.
This is the same caution FBWH's own self-hosted AI VPS comparison gives for any provider: keep Ollama on localhost unless you have a specific reason to expose it, and if you do, lock the source down.
Step 6: Snapshot Before You Experiment (Optional)
Before you start pulling different models or messing with configuration, it's worth taking a snapshot. If something breaks, you roll back instead of starting over.
- Go to Servers & Hosting → VPS → [your server] → More → Create Snapshot.
- Give it a name and an optional description, then click Create Snapshot.
A few things worth knowing about Contabo's snapshots specifically: they auto-delete after 30 days, they don't count against your VPS's own disk allocation, and they can't be downloaded — this is a rollback tool, not a backup export. It's only available on the Cloud VPS (High-speed series) line, which the Plus tier is part of.
Frequently Asked Questions
Can I run Ollama on Contabo's cheapest VPS plan?
Yes, for small models, with little headroom. Contabo's cheapest Core VPS, Cloud VPS 4, already has 8GB RAM (€5.50/mo list, €4.40 for the first 24 months), but it uses shared vCPUs and SSD storage.
This guide uses the Cloud VPS Plus 4 (4 vCPU AMD EPYC, 8GB RAM, 150GB NVMe, €13.50/mo list, €10.80 for the first 24 months) for faster storage and newer CPUs.
Do I need a GPU VPS to run Ollama on Contabo?
Only for larger models. This guide is scoped to small quantized models in the 7B-8B range, which run acceptably on CPU. For 70B-class models or production inference serving multiple concurrent users, Contabo's own GPU VPS inference guide is the right resource, not this one.
Is it safe to expose Ollama's port 11434 to the internet?
Not with an open Source field. Ollama's API has no built-in authentication, so anyone who can reach port 11434 can use your server's compute or read your model list. If you need remote access, restrict the firewall rule's Source to your own IP or a specific CIDR range, never "Any."
How fast will Ollama run on a CPU-only Contabo VPS?
Slower than GPU inference, but we haven't yet run a live benchmark on this specific plan to give you real tokens-per-second numbers. This section will be updated once that test is complete — treat any throughput figure you see elsewhere as an estimate until then.
What Ubuntu version should I choose when setting up my Contabo VPS?
Contabo's product page lists Ubuntu as an OS option without pinning a specific point release, so the exact versions offered (22.04 vs 24.04, for example) depend on what's shown in your panel at install time. Pick the newest LTS release available in that list.
Bottom Line
A Cloud VPS Plus 4 gets you a working, self-hosted Ollama setup for a 7B-8B model at €13.50/month list (€10.80 for the first 24 months), no GPU premium attached. The order flow, panel navigation, SSH key setup, and firewall configuration in this guide are all verified directly against Contabo's own documentation.
Related: Contabo review — every product line, pricing, and terms. | Best VPS for AI — Contabo, Hostinger, Hetzner, Vultr & DigitalOcean compared. | Self-Hosted n8n Hosting — 3 VPS Options Compared.
Data Sources
- Order flow, checkout steps, password policy: Contabo Help Center — help.contabo.com/en/support/solutions/articles/103000394299-how-to-order-a-contabo-product
- Cloud VPS Plus plan table, pricing, OS options: Contabo official product page — contabo.com/en/vps-performance/, verified September 2026
- Panel navigation, reinstall flow, SSH-key preload field: Contabo Help Center — help.contabo.com/en/support/solutions/articles/103000271913-how-do-i-install-my-contabo-server-
- SSH key generation and manual deployment commands: Contabo Help Center — help.contabo.com/en/support/solutions/articles/103000271398-how-do-i-set-up-an-ssh-connection-
- Firewall panel click-path (Network Services → Firewalling → rule creation → VPS assignment): Contabo Help Center — help.contabo.com/en/support/solutions/articles/103000390430-firewall-what-is-it
- Snapshot creation, retention policy, product-line restriction: Contabo Help Center — help.contabo.com/en/support/solutions/articles/103000270385-how-do-i-create-a-snapshot-of-my-server-
- GPU VPS inference guidance (link-out for larger models): Contabo Blog — contabo.com/blog/how-to-run-llm-inference-on-a-gpu-vps-2026-guide/
- Ollama install script: Ollama official — ollama.com/install.sh