30-SECOND SUMMARY
What to take away
- Use the official installer for your operating system.
- Check CPU/GPU loading with `ollama ps` before diagnosing performance.
- Local and `:cloud` models have different data paths; verify the model name.
When generation feels slow
Change one condition at a time.
- 01REPEAT
First run or every run?
- 02PROCESSOR
Check ollama ps
- 03MEMORY
Close heavy apps
- 04MODEL
Try one size smaller
Prepare the computer
Confirm the operating system, CPU architecture, free storage, and installation permissions. Model downloads can consume many gigabytes even when the Ollama application itself is small.
Use the official download page and connect a laptop to power before the first large download.
Install and verify the service
On macOS and Windows, run the official installer and confirm the Ollama application is active. On Linux, follow the distribution-specific official instructions.
Open Terminal or PowerShell and run the command below. If the service is not running, `ollama serve` exposes the startup error directly.
# Verify the CLI
ollama
# Start the server manually when needed
ollama serveRun the first model
The first run downloads the model and then enters chat mode. Start with a short question that is easy to verify. Exit with `/bye`.
The example model name may change over time, so confirm availability in the official model library.
ollama run gemma3:4b
>>> Explain local AI in three sentences.
/byeConfirm where the model is loaded
`ollama ls` lists downloaded models. `ollama ps` shows models currently in memory and the processor split. A model may run partly on the GPU and partly on the CPU when VRAM is limited.
If a supported GPU is not used, check drivers, other applications using VRAM, and whether the model is larger than available GPU memory.
ollama ls
ollama psTroubleshoot in a fixed order
Separate model-loading delay from ongoing generation speed. Close memory-heavy applications, start a new conversation, and compare a smaller model with the same prompt.
For download failures, check storage, proxy configuration, certificates, and official log locations before reinstalling.
- Compare first and repeated runs
- Inspect PROCESSOR in `ollama ps`
- Reduce context length
- Try a smaller quantization
- Read the official troubleshooting logs
Frequently asked questions
Why is only the first response slow?
The model may be loading from storage into memory. Repeated requests can be faster while it stays loaded.
Does `ollama stop` delete the model?
No. It unloads the model from memory. Use `ollama rm` to delete the stored model.
Can I expose port 11434 to the internet?
Do not expose it without authentication, network controls, TLS, rate limits, and a security review.
Primary sources
Check the original documentation for version-specific details.
Ollama Quickstart Ollama CLI Reference Ollama FAQ