They die because over time the filament sinters causing grain growth and necking at the edges. The individual grains shift until due to the 1D constraint they each have grain boundaries with each other that stretch from the radial edges of the cylinder across. Once the grain boundary looks like that, you see increased surface diffusion at the grain boundary near the surface that forms a pit which weakens the filament.
Perhaps you could argue that it requires cycling because the cooling causes increased axial stress and gradually pulls the filament apart, but it is most certainly more complicated than just thermal shock. I believe that so long as you're not shattering glass, it's more about how long it's been hot so far and how many times it's been completely cooled.
