I've just read this article and don't quite understand how a power supply can't supply such a low current. I was always under the impression a power supply could supply anything up to its rating, but didn't have a lower limit.
So why would they need to be redesigned?
From the article:
As it appears, Haswell's C6/C7 states require a minimum load of 0.05A on the 12V2 rail, and many desktop power supply units (PSUs) just cannot provide that low current, reports The Tech Report web-site. Meanwhile, numerous older PSUs, which comply with ATX12V v2.3 design guidelines only called for a minimum load of 0.5A on the CPU power rail, hence a less sophisticated internal feedback loop/protection could be used, reports VR-Zone web-site. As a result, unless C6/C7 power states are disabled in the BIOS, PCs with older/cheap PSUs may become unstable when processors enter these states.
A minimum load specification signifies the smallest load that can be drawn from the power supply while meeting all of the other requirements in the specification (regulation, transient response, etc.)
The power supply may or may not be able to deliver less current than what is specified as its minimum. It may deliver but drift out of voltage regulation; it may become unstable and oscillate; it may hiccup on and off; it may even go into overvoltage protection and latch off. Because the load is outside the specification, "anything goes".
The article's statement "just cannot provide that low current" is (to me) a gross simplification of the matter, and is a bit misleading. Current power supplies were never designed to meet this specific condition, so behaviour at this condition is undefined.
It is easier in some ways to design a high-efficiency switching regulator if you can assume that it has both a minimum load as well as a maximum load, reducing the "dynamic range" it must handle. Many PC power supplies are designed this way, both the main supply for the box, as well as on-board regulators for the CPU and memory.
The new chips violate the assumptions built into many existing systems and those systems cannot support the low-power modes without going "out of regulation" in some way, no longer meeting their specifications.
You could "work around" this issue by adding a "dummy load" (resistor) to the CPU power bus, but this would miss the point of having the low-power modes in the first place. It's simpler to simply disable those modes in software (the BIOS).
Switched mode power supplies work by transferring energy in pulses from input to output. With many topologies the duty cycle of these pulses must be reduced to very small values when under light load to maintain the correct output voltage. Some controller designs only operate over a limited range of duty cycles and therefore cannot maintain correct voltage when underloaded. This in turn may cause the power supply to trip out completely or to swing wildly between undervoltage and overvoltage.
Since there is a minumum practical duration of a pulse, power supplies that support zero minum load will typically reduce the duty cycle by increasing the delay between bursts when under light or no load. This is why some power supplies buzz when under light or no load. The increase in delay between bursts reduces the switching frequency into the audible range.
There are also some designs where pulses of energy can move in both directions between the input and output side due to the use of a second transistor rather than a diode. These avoid the need for very low duty cycles under light load but typically have higher no-load power consumption.
Its not so much that they "can't" as much as it is they are just not designed to. Computers use switching regulators that are designed to operate around a certain current and voltage but do not perform well if the current and voltage deviates too much from this point.
