A tall support tower topples partway up, usually taking the overhang it was holding with it. The nozzle then either drags the fallen material around or prints into thin air over the unsupported region.
Why a support tower is a bad column
Think about what the slicer has asked for: a tall, thin, low-density column of plastic, unconnected to anything else, built one soft layer at a time.
Three things work against it. Each nozzle pass applies a sideways force at the very top, where the leverage is greatest. Each new layer arrives molten, so the top of the column is always the weakest part of it. And the footprint is small, because supports are deliberately minimised to save material and time.
Beyond a certain height-to-width ratio, the sideways force from printing wins. That is not a settings failure so much as a structural inevitability, and it is why the fixes below are mostly about geometry rather than about tuning.
Fixes, in order of how much they change
Switch to tree or organic supports. This is the structural answer rather than a workaround. Tree supports brace themselves against the model as they rise, so the free-standing height of any individual branch stays short even on a tall part. They also use less material and touch the surface at fewer points. On a part whose supports keep falling over, this single change usually ends the problem.
Turn on a support brim. Extra bonded area at the base, exactly as a brim does for the part itself. Cheap and effective for towers that are toppling rather than buckling.
Raise support density. More internal lines per layer means each layer is stiffer and better connected to the one below it. Somewhere around 20% is a reasonable starting point for conventional supports; below about a tenth they are fragile by construction.
Reduce travel acceleration. Some collapses are caused by the toolhead sweeping past at speed rather than by the printing pass itself.
Reorient the model. The support that never has to be tall cannot fall over. A part rotated so its overhangs are closer to the plate needs shorter towers, and often fewer of them.
The special case of an isolated island
A tower supporting a small feature high up on a model — the tip of an outstretched arm, the top of a spout — is the worst version of this. It is tall, thin, unconnected to anything, and it carries almost no load, so it gets the minimum density the slicer will give it.
For those, tree supports are strictly better. Failing that, manually painted supports at higher density in that region only, or a small design change that connects the feature to the body, will do more than any global setting.
What it looks like after the fact
A collapsed tower is not always obvious in the wreckage. What you often find is a heap of loose material at the base and an overhang above that printed into air — which reads as a spaghetti failure unless you look for the fallen tower in the mess.
If a print failed high up and there is a stub of support left standing at the bottom, that is the signature. It is worth identifying, because the fix for a collapsed support is nothing like the fix for a detached part.
Do not fix it by adding supports everywhere
The reflex is to increase support coverage until nothing is unsupported. That multiplies material use and print time, leaves more scarred surfaces to clean up, and does not address the collapse — a hundred fragile towers fall over exactly as readily as ten.
Fewer, better-braced supports beat more supports every time.
Materials that make towers fragile
Anything with poor layer bonding builds weak towers, so silk and matte PLA grades collapse more readily than they should for their stiffness. TPU is hopeless in this role because the tower flexes rather than resisting. Fibre-filled materials are the opposite — the stiffest supports here, and they break away more cleanly too.
High-contraction materials add their own twist: ABS and ASA support towers can warp away from vertical as they cool, so a tower that has not fallen over may still have leaned far enough to be useless where it was needed.