Why are server power supply unit (PSU) ratings so much higher than the actual server power consumption? In Part 1 of this blog, I analyzed some server data and concluded that while the diversity of server power consumption is greater for new servers, it doesn’t suggest that this is the main reason for oversized PSUs ratings. In Part 2, I look at another reason, and likely THE reason for this oversizing: Economies of scale. This is an economics principal that all the major manufactures (including server vendors) use to lower their cost of goods.
If the economies of scale principal didn’t exist, then theoretically a manufacturer could produce a single 500 watt PSU (or 15A circuit breaker) for the same cost as 1,000, or even 1 million. Furthermore, they could buy small batches of 10,000 different PSU capacities to rightsize almost every server. But clearly anyone who has bought in bulk knows this not to be the case. And anyone who tries to buy a 7.25 amp circuit breaker for their breaker panel won’t find one.
Here in lies the main reason why a server with a maximum power consumption of 220 watts doesn’t ship with a 220 watt PSU. It is more economical for server vendors to oversize power supplies and carry a small number of power supply SKUs (thereby benefiting from economies of scale) than carrying more SKUs that allow better right-sizing. I approached a former employee from a server vendor and he confirmed this is the case.
He also mentioned two other factors that have bolstered this oversizing practice. One is that the 80Plus power supply standard has improved PSU efficiency curves so that even I light loads, the PSU efficiency is relatively high. The second factor is that PSU costs have decreased over the years making server vendors less sensitive to oversizing PSUs.
So does this mean that we’re stuck with this paradigm of paying for 3 to 6 times more PSU capacity than we’ll ever need? Not if the Open Compute Project (OCP) has anything to do about it. The OCP power architecture uses a centralized PSU for a single rack, similar to how blade servers share the same PSUs within a blade chassis. In essence, the entire OCP rack is a blade chassis. Schneider’s Data Center Science Center and Systems Engineering group recently compared the capital cost between a traditional data center power architecture and an OCP power architecture and found that centralizing PSUs represented a material savings. This savings is likely to be higher if OCP rack adoption grows due to economies of scale. For more information on this analysis, see White Paper 228, Analysis of Data Center Architectures Supporting Open Compute Project (OCP).