This got me curious so I looked up some runtime graphs. Here’s what I found:
Zeroair did some nice measurements…
It appears to have an unregulated DD FET driver with no “+1” regulator for low modes, so even the low modes have sag as the battery drains. It also has one of the most unstable thermal regulation patterns I’ve ever seen.
This got me curious too, so I looked up graphs for one of those lights too. The first one I found was a Wurkkos FC13, also by Zeroair. Due to its similar size and similar FET-style driver for high modes, it can sustain a similar brightness level when limited by temperature:
This light uses a FET+1 driver so it also doesn’t have great performance, but at least the low modes (under ~130 lm or so) are regulated for flat output… and that’s what you’d be using if you want it to run for 8 hours. The “+1” chip (amc7135) keeps output almost completely flat. Those modes weren’t runtime-tested though, since the test would take a long time and the performance is already pretty well understood from testing other lights with the same chip.
The FC13 thermal regulation isn’t the most stable, since it’s constantly fighting against FET sag and measurement noise, but it at least stays within about 10% of the optimal level, instead of swinging wildly back and forth. For reference, 7% is the threshold Cree uses for its binning process, since people can’t generally perceive a 7% difference in brightness… so the regulation here is mostly invisible after it reaches the desired temperature.
But a FET+1 driver isn’t great. It’s simple and cheap so it gets used a lot, but one could do a lot better.
For example, when using a FET in linear mode, output can be much more stable. This method still burns off extra power instead of converting it to more lumens, but at least it keeps the output steady:
The increases toward the end of the turbo+fan run are because this type of driver gets more efficient when the battery is low, generating less heat… so it turns up the brightness to compensate.
… and then there’s boost / buck drivers. The graphs for those look the same as the linear driver, except they can maintain higher output at the same temperature, and they get longer runtimes at the same brightness.