These overclocking problems have a common source: VRMs. Fundamentally, VRMs are power supply units. The conception of such devices is a difficult work (voltage and current stability, size of heatsinks, OCP (over current protection), etc.) and sometimes some parts may be under-sized. Under-sized VRMs can lead to a malfunction of the card when the current the VRMs have to deliver is too high (due to a high GPU load). Recent cards like Radeon HD 5000 series come with an advanced OCP management (the GPU is throttled back to reduce the current the VRMs have to deliver, see HERE) but on other cards, an over-current may lead to a system reboot (maybe due to the rig PSU OCP too) or even kill the card (see HERE – in that case, the OCP didn’t work at all…). In this article, I’ll try to explain the basic notions about VRMs and where they are located on the PCB (printed circuit board) of a HD 5970.
These principles are more or less the same for all graphics cards (the components can be different but have the same function).
As usual, if you have additional information and resource or if you detect some errors, just drop a comment.