“Engineers designing PDNs for densified computational resources are challenged to deliver more power to compact, power-hungry
processor cores without overheating system components.”
A key factor driving the reduction in size of electronic components and data systems is demand for greater cost efficiencies in processing an exponential growth in data. All that data is collected, stored, manipulated, and analyzed. In the data business, it’s all about data per square foot. If you can pack more processing power into the unit volume of a building, then the cost of your data service goes down.
Engineers designing power distribution networks (PDNs) for densified computational resources are challenged to deliver more power to compact, power-hungry processor cores without overheating system components. That requires thinking about PDN design in different ways, not just replicating old designs because they worked in the past.
For example, a typical server will have power coming in at 48 volts. Inside that server are processor cores doing all the calculations, which run at about 1 volt. Those processor cores on the 1-volt rail are consuming the bulk of the power in the server. The flow of energy is not steady, either, because sometimes the processor cores are working hard and drawing perhaps hundreds of amps, and then they pause and use little power. Sometimes, they operate at medium capacity. Their operating power
requirements can be highly variable under normal conditions.
This is an excerpt from 7 Experts on New Approaches for Power Distribution
Network Design. The eBook was generously sponsored by KEMET Corporation and Mouser Electronics.