Is 100% CPU Usage Normal for Ollama?
ollama ps shows the PROCESSOR column pegged at 100% CPU during a response. Usually yes, that's expected - here's how to tell if it actually isn't.
Short answer: on a CPU-only setup, yes
If you're following the mini PC guide, there's no GPU in the machine at all - the CPU is doing 100% of the work by design, not because something's wrong. Seeing 100% CPU in the PROCESSOR column of ollama ps while a response is generating is exactly what a healthy CPU-only setup looks like.
When it's a different situation: a GPU build
If you're on the full PC build with a discrete GPU, 100% CPU in that same column means the opposite of normal - it means the GPU isn't being used for this model at all. That's a real, different problem, covered in Why Isn't Ollama Using My GPU?.
How to tell if a CPU-only setup actually has a real problem
100% CPU alone isn't the symptom to watch. These are:
- Responses are painfully slow for the model size you're running - almost always a RAM problem, not a CPU one. See the mini PC guide's troubleshooting section.
- The system is swapping to disk while generating - check memory usage (Activity Monitor on Mac,
htopon Linux) alongside CPU. Swapping, not CPU load, is what actually tanks performance. - Sustained thermal throttling - a mini PC under sustained 100% CPU for long sessions can throttle if airflow is poor. If performance degrades the longer a session runs rather than staying consistent, check the machine isn't overheating.
If none of those apply and responses are simply taking the amount of time you'd expect for a CPU-only setup at that model size, there's nothing to fix - that's the actual, correct behavior.
Written from hands-on security operations experience. More about this site →