The data center industry’s growth projections can be met by combining energy supply growth and demand reduction. Highly utilized IT infrastructure and efficient software can mitigate demand growth while delivering needed IT capacity.
Hyperscalers design their own servers and silicon to scale colossal server estates effectively. AWS uses a system called Nitro to offload virtualization, networking and storage management from the server processor onto a custom chip.
If adopted, the UNEP U4E server and storage product technical specifications may create a confusing and counter-productive regulatory structure. The current proposals are as likely to limit as improve data center operations' efficiency
Dedicated GPU infrastructure can beat the public cloud on cost. Companies considering purchasing an AI cluster need to consider utilization as the key variable in their calculations.
Supersized generative AI models are placing onerous demands on both IT and facilities infrastructure. The challenge for next-generation AI infrastructure will be power, forcing operators to explore new electrification architectures.
Nvidia’s dominant position in the AI hardware market may be steering data center design in the wrong direction. This dominance will be harder to sustain as enterprises begin to understand AI and opt for cheaper, simpler hardware.
Visibility into costs remains a top priority for enterprises that are consuming cloud services. Improving the tagging of workloads and resources may help them to spot, and curb, rising costs.
Not all generative AI applications will require large and dense infrastructure footprints. This complicates AI power consumption projections and data center planning.
Many organizations still do not tap into the potential power efficiency gains hidden in servers. Without operational focus on extracting those, future server platforms may bring marginal, if any, energy performance improvements.
Enterprises have various options on how and where to deploy their AI training and inference workloads. This report explains how these different options balance cost, complexity and customization.
Although quantum computing promises a revolution in scientific discovery, its use is still constrained to research and continuing development. However, a new IBM quantum data center in Germany signals a growing interest in its capabilities.
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?
Jay Dietrich
Dr. Owen Rogers
Daniel Bizo
Max Smolaks
Dr. Tomas Rahkonen
John O'Brien