Brochure
Powering data centers from grid to chip
Driven by insatiable demand for cloud-based storage and processing, data centers are expanding at an unprecedented rate. The total worldwide capacity of today’s hyperscale data centers, the largest types, has doubled in the last four years, and continues to grow.
This expanding data center capacity feeds an ever-increasing demand for more power. Components need to be reliably supplied with energy, and efficiency remains a top priority. At the same time, there’s an increasing need for compact systems with higher power density that can work with liquid cooling systems with reduced heat dissipation to reduce the cost of thermal management.
The shift to 48 V
In response, one of today’s trends is a shift away from 12 V as the standard to distribute DC power within the data center, and to instead use 48 V. Increasing the voltage fourfold reduces the required current by the same factor. This means that conduction losses are reduced by a factor of 16 (as they are proportional to the square of the current), thus improving efficiency and reducing the heat dissipated, while busbars can be smaller.
At the same time, data centers are still full of systems and components that need a well-regulated 12 V supply, where the output voltage of the power supply doesn’t fluctuate with variations in input voltage. Examples of components needing regulated 12 V supplies include PCIe cards, certain memory types, disk drives or SSDs for storage, and cooling fans.
In practice, the input voltage can often vary in a data center, and not all power system topologies are able to cope with this. This is reflected in chamber tests by the Open Compute Project (OCP), where varying and noisy input voltage is introduced.
Hyperscale data centers need modular, scalable power systems that can be deployed quickly if needed — for example, if a customer moves equipment around in their racks, or adds new computing or storage capacity.
