Doubling is exactly what the description says. Doubling basically means adding components in parallel / splitting the load between the components. When done properly (using doublers) doubled VRM is absolutely fine. Increasing the number of phases either by adding more native ones or through doubling will improve (reduce) the power / power dissipation density, which is always a good thing. Ripple will also reduce when the phase count goes up. It also helps to get around component limitations like maximum current, much like using two cables in parallel to safely carry 15A when a single cable can carry 10A of current. Despite doubling the hardware will reduce the per component load, most if not all VRMs are limited by the total power dissipation and not the ratings of a single component.
Some of the doubled solutions using lower-end VRM controller with fixed or otherwise limited switching frequency will end up being lower quality than a native VRM, regardless if the doubling is done properly or not. That's because doubling will split the PWM pulses in two and the effective switching frequency will be half of the original, or a quater of the original in case of quadrupling. On the higher-end controllers (CHIL / IRF or Infineon) that is not a problem since the switching frequency in controller side can be raised high enough, to provide sufficient effective switching frequency. Some of the manufacturers (without naming anyone specific, the usual suspects) have been doubling the phases without using the proper hardware for the job. They have simply connected the hardware of those phases into the same gate driver and controller PWM output (fake phase).
Both 970A-UDP3 ("8+2 phase" / doubled 4+1 phase) and 970A-D3P use the same IR3564B VRM controller. It is a 4+1 phase controller which means that it has four PWM outputs for the Loop 1 (VDDCR, CPU) and a single PWM output for Loop 2 (VDDNB, NB). Each of the PWM output of the controller is connected to the PWM input of the gate driver. Each gate driver controls the mosfets of a single phase, or two phases if the driver happens to be a double driver. This is the definition of a native phase configuration (i.e each phase has it's own PWM output in the controller). A properly doubled solution has one additional component per two phases, the doublers. The doublers are placed between the gate drivers and the VRM controller. A single PWM output from the VRM controller enters into the doubler, where the pulses are split in two (e.g. on UD3P it would be 90° to 45°). The doubler has two PWM outputs which are then connected to the gate drivers, same way as on the native solution.
Edited by The Stilt - 6/5/16 at 6:24am