Keeping Older Hardware Cool

Local inference defaults to all cores plus max turbo: on old laptops that means 90 C, throttling, and jet-engine fans. Three dials fix it.

1. App-side thread limit

In ~/.config/qwen-desktop/config.json:

"ollama_threads": 2

2. Pin Ollama to two physical cores

Check topology (0,4 means siblings sharing one core):

cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list

Then run:

sudo systemctl edit ollama

and add:

[Service]
CPUAffinity=0 1

then apply:

sudo systemctl daemon-reload
sudo systemctl restart ollama

3. Cap turbo (optional)

sudo cpupower frequency-set -u 2200000

Persistent across reboots: edit /etc/default/cpupower and set max_freq="2.2GHz".

Result

2 threads + 2 cores + 2.2 GHz cap = silent fans and 65-75 C during generation on a 2012 i7.