Understanding the power requirements of an individual embedded computing card is an important part of overall systems design. What the card will require from the system power supply and, ideally, how those requirements will vary based on workload, is information that factors into system design decisions. Obviously there must always be sufficient power for all system components, across all required voltages and individual power rails. Determining the power needs is the first step.
A very traditional engineering approach is to make a best estimate of the power requirements for each card in a system using whatever information is available, usually from vendor spec sheets. It is probably an exaggeration to call this guesswork, but not by much. Next, sum those requirements across all the cards and, finally, apply a safety factor. This “safe” number is then used to define the capacity of the system power supply.