Posted on September 9, 2021 by Neche Veyssal
In this blog from Raritan, we address the new ways that academic researchers have to learn how to support their supercomputing applications.
Through the most natural forms of osmosis and attrition, our public institutions of higher learning seem to have stepped into the role of ‘sun within the perceived galaxy’ known as supercomputing. Perhaps by definition, the supercomputer has evolved to a computation-performance level that, despite being man-made technology, reaches far beyond the capabilities of mere human reason or intellect. In as much as they help to expand the limits of our understanding, these monoliths also push the envelope in terms of the physical support that they require, casting academicians and students into roles meant for hyperscale or colocation-facility managers.
Let’s look at three areas that demand attention when supporting higher densities in higher-education applications:
Supporting More kW
As the load edges past 15 kW per rack, it becomes increasingly difficult and more expensive to provide circuits to each rack to meet demand. For example, it would take two separate 30-amp circuits at 208V to meet the load, while only one circuit would be required at a higher voltage. Double these numbers for redundant A+B power, four circuits versus two, and you can imagine the difference in cost and complexity for supercomputing applications.
Raritan rack PDUs can be configured and shipped to support 415V. Although most often applied to greenfield applications, the ability to radically increase the amount of power that is ultimately delivered to the rack can also be made justifiable in brownfield situations. Distributing power to the rack at the highest voltage possible not only provides the most economical means of powering higher densities, but it is the most ecological, too.
One environmental advantage pertains to electrical losses. At the device level, supercomputing arrays and network appliances operate from 4–6% more efficiently at 415V than 208V. And while this is significant, additional savings come through the distribution system that brings power from the entrance and ultimately to the rack. The number and degree of losses due to transformation through UPSs, PDUs, and RPPs can add an additional 2–3% efficiency gain to the system.
All of this adds up—distribution and transformer losses alone can rack up $21,900 per year at loads of 500 kW (assuming $.10 per kWh). The savings that come from these efficiencies multiply as the total IT load increases.
Intelligent PDUs, or rack-power-distribution devices with built-in networking, monitoring, and data-collection features, can prove indispensable in supercomputing applications. Intelligent rack-power-distribution equipment can be leveraged to help keep systems secure by providing input at the rack level of critical power-, temperature-, and environmental-monitoring points—to ensure that the system is protected.
Intelligent PDUs can also tie to sensors such as contact closures to ensure this equipment is also physically secure. This allows operational and physical security-monitoring points to be accessed directly from the devices, through secure software, or through campus BMS and DCIM software.
At Legrand, we are committed to supporting higher education and Research I facilities with intelligent rack power distribution from Raritan. To learn more about how to support higher kW loads while improving sustainability and security, check out our Environmental and Security Solutions pages.