25 tok/s
7B params
Device
45C
Actual Tokens/s
25
Power Draw
24W
Status
Nominal
Time to Throttle
--
Temperature History
95C
70C
Sustained Performance
| Device | 7B tok/s |
|---|---|
| M3 Air (fanless) | ~12 |
| M2 Max (laptop) | ~35 |
| RTX 4090 (desktop) | ~80 |
| H100 (datacenter) | ~150 |
Physics
Thermal throttling begins at 95C. Fanless devices rely on passive cooling (~0.5W/C dissipation). Active cooling provides ~3W/C. Sustained performance depends on thermal headroom, not peak capability.