Balance is the new challenge for datacenters designers

Many threats to the continued development of datacenters will loom in 2023 and beyond. While the digital infrastructure sector has grown rapidly in recent decades and demand for compute and storage resources is stronger than ever, global market conditions are more difficult today. In financial terms, they will be the insistent high price of energy, costs of building and operating these environments and tight budgets. Among other challenging elements, we will also continue to see shortages of skilled labour, climate change issues and tighter sustainability standards.

However, this does not mean that there will be a slowdown in the data centre industry, quite the contrary. The explosion in demand for capacity in these environments has attracted the attention of more investors in several categories, including equity, buyouts, real estate and, increasingly, infrastructure investors. In the United States alone, demand, generally measured by energy consumption, is expected to reach 35 gigawatts (GW) by 2030, up from 17 GW in 2022, according to McKinsey. The United States accounts for about 40% of the global market.

According to a Uptime Institute survey forecasting 2023, the majority of data centre owners and operators surveyed recently plan to expand their footprints this year to meet rising capacity demand. However, data centre operators will begin to face a dilemma: how to balance increasingly powerful processors while reducing power consumption?

The servers used by data centres have had relatively stable power and cooling requirements for nearly two decades, and with this technical stability, it has been possible to design installations and retrofits with some peace of mind, helping to attract investment.

That appears to be changing. Uptime Institute research has shown a consistent trend of increasing data centre rack power density over the last decade. Contrary to some more aggressive expectations, the power consumption of a typical rack has remained below 10 kW. However, this scenario has picked up pace again more recently, and the trend is continuing. And, according to the study, the increase in power density is not exclusively due to more overloaded racks. It is also due to higher power consumption per server, which is mainly being driven by the mass emergence of servers with higher power processors, attractive for their high performance and often better energy efficiency if well-used.

So while data centre designers were once able to reliably and consistently plan the power and cooling density of facilities based on a stable and predictable consumption per processor and other server components, today that calculation is much more complex. The accelerating power density of IT resources has made it difficult to make plausible design assumptions for environments that need to be technologically resilient and sustainable over several years or even decades.

And denser processors are already a reality. Models from giants like Intel and AMD, consuming up to 400W, are already circulating around. 4th generation Intel Xeon Scalable processors have thermal design power (TDP) of up to 350 watts, while AMD’s 4th generation Epyc processors have TDP of up to 360 watts. And we may see new data centre processors with TDPs of up to 600 watts in the coming years, making it even more difficult to establish design assumptions for these environments.

This new scenario increases technical and business risks. Additional costs due to incorrect calculations or assumptions in data centre design can be significant. More than ever, it is necessary to find the right balance – on the one hand, taking a very conservative stance, such as low power density approaches, can cause the data centre to quickly become constrained or even obsolete; on the other hand, being technically aggressive, assuming high-density racks and heat re-use, can lead to overspending on capacity and under-utilised resources.

“In short, facilities built today need to remain economically competitive and technically capable for 10 to 15 years. This means that certain assumptions must be made through speculation, without designers really knowing the future specifications of IT racks. Consequently, engineers and decision-makers must deal with uncertainties surrounding the technical requirements of data centres throughout the second half of the 2020s and beyond,” Uptime Institute points out.

Today, even the best-informed can only speculate, the study concludes.