Benchmarks
Tokens per second, watts, and tokens per watt for every mini PC we've run. Latest result per machine wins. Methodology →
llama3.1:8b
| # | Machine | tok/s | watts | tok/s/W |
|---|---|---|---|---|
| #1 | Aoostar GEM12 backend: gpu | 8.3 tok/s | — | — |
llama3.2:1b
| # | Machine | tok/s | watts | tok/s/W |
|---|---|---|---|---|
| #1 | Aoostar GEM12 backend: gpu | 48.1 tok/s | — | — |
llama3.2:3b
| # | Machine | tok/s | watts | tok/s/W |
|---|---|---|---|---|
| #1 | Aoostar GEM12 backend: gpu | 26.7 tok/s | — | — |
qwen2.5:14b
| # | Machine | tok/s | watts | tok/s/W |
|---|---|---|---|---|
| #1 | Aoostar GEM12 backend: gpu | 6.2 tok/s | — | — |