AI training clusters can show rapid and large swings in power consumption. This behavior is likely driven by a combination of properties of both modern compute silicon and AI training software - and may be difficult to manage at scale.
Increasing supply air temperature is gaining interest as an approach to potentially save data center energy. However, savings will not be universally possible and understanding its potential involves a complex multivariable analysis.
Pulling reliable power consumption data from IT is increasingly important for operators. Although third-party software products offer promise, significant roadblocks still hinder adoption.
Most operators will be familiar with the outrageous power and cooling demands of hardware for generative AI. Why are these systems so difficult to accommodate, and what does this mean for the future of data center design?
Several recent outages have exposed the global dependency on a small number of third-party suppliers - and governments around the world are already taking note.
Metered-by-outlet iPDUs present a relatively straightforward method of collecting server-level power consumption data. This information will be increasingly important to data center efficiency - making iPDUs a more popular choice.
Avoiding digital infrastructure failures remains paramount for data center owners and operators. This report analyzes recent Uptime Institute data on IT and data center outage trends: their causes, costs and consequences.
Industry stakeholders recognize that to truly understand IT infrastructure efficiency, data center operators need to report a facility work per unit of energy metric. Most operators are, however, unprepared to calculate this metric.
There is more to managing server power than just conserving energy when the machine is idle. Another side to optimizing energy performance involves setting processor performance levels appropriate for the application.
Without the active contribution from IT, data center infrastructure energy performance and sustainability will fall short of aspirations. Server power management features remain unexplored and underused by most efficiency initiatives.
Operators are missing opportunities to lower costs and energy use by not using utilization and power management data, an Uptime Intelligence survey on IT and power efficiency suggests.
Data center and IT managers face growing demand to publish comprehensive carbon inventory reports. But estimates for the carbon content embedded in IT equipment have questionable accuracy and usefulness in informing decisions.
Pressure to improve data center efficiency and sustainability is driving interest in artificial intelligence (AI) technologies. Several startups aim to deliver new capabilities in IT power management and cooling optimization.
Results from the Uptime Institute IT and Power Efficiency Survey 2023 suggest that many organizations may struggle to meet regulatory requirements for energy, carbon and IT life cycle tracking.
Despite high expectations, most operators will only see moderate impact from specialized AI hardware installations in the immediate future. The emergence of AI as a major force will sway the industry in a more profound, but less direct, fashion.
Daniel Bizo
Dr. Tomas Rahkonen
John O'Brien
Max Smolaks
Andy Lawrence
Douglas Donnellan
Jay Dietrich
Rose Weinschenk