I purchased a Mac mini with 32GB to try running OpenClaw.
I tried running it with ollama(qwen3-30b-64k) + OpenClaw (bare bones, no features installed)... and the performance is no joke.. ㅋㅋ
This is roughly this level..
I entered 'hi' and 'hello' in the OpenClaw TUI, but it took about 10 minutes to get a response.
(At first, I thought I had configured it incorrectly.)
Memory usage was 30GB, and swap was around 6GB
So I checked ollama only
ollama run qwen3-8b-64k:latest "hi" : This takes 5-10 seconds
ollama run qwen3-30b-64k:latest "hi" : This takes 20-30 seconds
Memory usage was between 20-27GB..
I understand that OpenClaw passes various additional contexts together.. but it's doing too much..
So I gave up on this combination.
I'm going to change to using a commercial API and set it up again to try it out..
The performance was much worse than expected, so I gave up right away. This doesn't seem like a problem that can be solved by tuning or configuration.. ㅎㅎ
I will organize the details in a blog post later and include only the key points here.
