Skip to main content

Snapshot prices 2026-05. Check on hetzner.com/cloud for current pricing. AWS and DigitalOcean prices are EU on-demand estimates. This tool does not constitute purchase advice.

Configure the workload

Expected requests/second for the app (excluding LLM inference).
OS + Runtime + Database + Cache + Application, without LLM
Includes OS, LLM models (llama3.1:8b ~5 GB), DB, logs
Hetzner includes 20-60TB inclusive for each Tier.

LLM hosting models (on-premise Ollama)

Select the models to run concurrently. The memory is summed up.

Recommended Tier

CCX13
4 AMD vCPUs 16 Gigabytes of Random Access Memory (RAM) 240 Hard Disk Drive (HDD) vs Solid State Drive (SSD): What's the Difference? 20 Including tax
€23.00/month
Yearly€276.00

Why this layer?

Application RAM4 GB
Large Language Model (LLM) architectures6 GB
Total Memory (RAM)10 GB
Hold for thirty percent buffer time+3 GB
Total requested RAM13 GB
Required storage space50 GB

Scaled for RAM (app + LLM + 30% buffer)

Comparison Providers Equivalents

vs Amazon Web Services (AWS) On-Demand in the European Union (EU), specifically Frankfurt, and Digital Ocean's basic plan.
Instance/ ProviderMonthlyAnnualyHetzner vs AWS (Amazon Web Services)
Hetzner CCX13Recommended€23.00€276.00Reference
AWS t3.mediumAWS€30.00€360.00 +€7.00/mo
AWS t3.largeAWS€60.00€720.00 +€37.00/mo
AWS t3.xlargeAWS€120.00€1440.00 +€97.00/mo
DO Basic 4 GBDO€24.00€288.00 +€1.00/mo
DO Basic 8 GBDO€48.00€576.00 +€25.00/mo
DO Basic 16 GBDO€96.00€1152.00 +€73.00/mo
Savings equivalent to Hetzner vs AWS per year__:__month__*/year___€84.00

All Hetzner CCX AMD Dedicated vCPU Tiers

CCX23
8 vCPU · 32 GB RAM · 240 GB SSD
€36.00/mese
CCX33
16 vCPU · 64 GB RAM · 480 GB SSD
€72.00/mese
CCX43
32 vCPU · 128 GB RAM · 960 GB SSD
€144.00/mese

Note: all CCX tiers use dedicated AMD EPYC CPUs (not shared like CX shared). For high-competition GPU/AI workloads, consider Hetzner Robot with a dedicated GPU (not available in standard Cloud UI).

Come utilizzare VPS Hetzner Cost Calculator

Insert workload

Set RPS target, app RAM (no LLM), total storage and predicted monthly traffic. You can use precompiled scenarios "Scenario 3 LLM" or "API Production".

Choose models for hosting LLMs

Select Ollama on-prem models you expect to run concurrently (llama3.1:8b, qwen2.5:7b, mistral:7b, llava:7b). RAM required is added.

Read the recommended tier

The tool automatically selects the minimum CCX tier that meets total RAM, storage, and RPS requirements, displaying monthly and annual costs.

Compare with AWS and DigitalOcean

In comparison table verify estimated savings compared to AWS t3 and DigitalOcean Droplet with similar specifications.

Suggerimenti

  • Use the "Scenario 3 LLM" scenario to simulate a VPS hosting multiple Ollama models concurrently and immediately verify its impact on total RAM.
  • Plan for growth in traffic or storage - leave buffer space: switching from one tier to another on Hetzner requires a resize (brief downtime) and is not instantaneous.
  • For high-competition GPU/AI workload, Hetzner Robot with dedicated GPU is valued over cloud-based UI standard on which this calculator is based.

Domande frequenti

How is the Tier CCX recommended amount calculated?

Tool sums application and model LLM selected RAM, applies a 30% buffer for operating system and cache, then selects the first suitable Hetzner CCX tier (13/23/33/43) with sufficient RAM and storage. If no tier is sufficient, proposes the largest available (CCX43).

How much RAM do selected LLM models occupy?

Approximate memory usage: Q4_K_M - 6GB, qwen2.5 - 5GB, mistral - 5GB, llava (vision model) - 8GB. These are estimated memory requirements for Ollama on-prem execution, not GPU-accelerated.

Are prices shown up-to-date?

Prices are an outdated snapshot (visible in the box disclaimer above) and are updated periodically. Always check the current quote on hetzner.com/cloud before a purchase: AWS and DigitalOcean prices are indicative estimates for on-demand EU.

Why does Hetzner CCX compare favorably to AWS or DigitalOcean for similar workloads?

Concise and similar length translation: Be concise — keep similar length. Output ONLY the translation, no quotes, no explanation. Italian: I tier CCX use dedicated AMD EPYC vCPU (non shared) at a fixed monthly price, with traffic ranging from 20 to 60 TB per tier. AWS on-demand and burstable instances like t3 often cost more than equal to RAM, while DigitalOcean has shared CPUs on Basic plans.

What do the warnings below the recommendation mean?

Check conditions: LLM inference on CPU (latency 10-60s per token without dedicated GPU on CCX13), declared traffic over 5,000 GB/month, storage above 80% of included disk in tier, or requires dedicated GPU for high-volume AI workload (available only on Hetzner Robot, not Cloud UI).