Thermal output management is not new. If the chip overheats, you can always clock itself down or power down some cores. GPUs, being very parallel, should make it even easier than it is with CPUs.
Not being able do transparently do this should be considered an important design flaw.
Sure, but when your chip tends to overheat, you should design mechanisms to reduce the thermal output when the need arises. CPUs have similar mechanisms integrated in them since the early 2000s (remember the videos of AMD CPUs melting down seconds after the removal of a heatsink?).
Not being able do transparently do this should be considered an important design flaw.