Section
Self Host LLM on VPS
Everything on self host llm on vps: start here for tested guides, comparisons and calculators.
Buyer's guideVPS for Ollama: which server to rent, by model size (5 servers tested)We ran Llama 3.2 1B, 3B and Llama 3.1 8B in Ollama on five Hetzner servers. Measured tokens/s, RAM and cost per million tokens, plus which VPS to pick
Buyer's guideHost an AI agent 24/7: what it needs and which server to rentHow to host an AI agent 24/7: restarts, RAM and model choice sized from our n8n and Ollama tests on 5 Hetzner servers, plus untested hosts priced Nov 
ComparisonCPU vs GPU for LLM: when a cheap CPU server is enoughCPU vs GPU for LLM, with our measured tokens/s on five Hetzner CPU servers turned into reading speed and answer time, plus RunPod GPU prices checked N
Cost guideCost per million tokens, self-hosted: measured on five CPU serversCost per million tokens self-hosted, worked out from tokens/s we measured on five Hetzner servers, at 24/7 and lower use, against API prices checked N
BenchmarkOllama CPU-only performance: tokens/s, prompt speed, load time and RAM on 5 serversOllama CPU-only benchmark on five Hetzner VPS: generation and prompt tokens/s, load time and RAM for Llama 3.2 1B, 3B and Llama 3.1 8B, with exact com
GuideLlama 3.1 8B requirements: what it really needs on a CPU serverLlama 3.1 8B requirements, measured: it failed on two 4 GB servers, ran on 8 GB at 5.7 to 6.3 tokens/s and on 16 GB at 9.7 tokens/s. RAM, load time, Q
GuideLlama 3.2 3B requirements: measured RAM, speed and the server you needLlama 3.2 3B requirements measured, not estimated: system RAM, tokens/s and load time in Ollama on five CPU-only servers, plus the minimum and comfort
How-toInstall Ollama on a VPS: the commands we ran on Ubuntu 24.04, with timingsInstall Ollama on a VPS with the official script on Ubuntu 24.04: install and model pull times measured on 5 servers, firewall, and keeping port 11434
ExplainerOllama on an ARM server: Hetzner CAX vs x86 CX on the same modelsOllama on ARM server vs x86, measured: Hetzner CAX11 and CAX21 (Ampere) against CX23, CX33 and CX43 on Llama 3.2 1B, 3B and 3.1 8B. Speed, RAM, price In the Lab next
- GPU VPS for LLM2026-11-11
- Self Hosted LLM vs OpenAI API Cost2026-11-12
- Ollama Tokens Per Second Benchmark2026-11-14
- Llama 3.2 1B Requirements2026-11-15
- Qwen 7B Requirements2026-12-02
- Mistral 7B Requirements2026-12-03
- Gemma Requirements2026-12-04
- Qwen3 8B Requirements2026-12-05