Cutting-edge server chips are turning heads for their potential to boost performance, but next-generation processors also run hotter than older designs, and data centre operators will struggle to figure out what to do with limited guidance from chip makers.
At the same time, there’s going to be increased scrutiny on the role that IT equipment can play in energy efficiency efforts. These interrelated trends are among the top predictions Uptime Institute is making for data centres this year.
“Operators will struggle with new, hotter server chips,” said Jacqueline Davis, research analyst at Uptime, during a web conference on the institute’s 2023 data centre predictions. Meanwhile, “energy-efficiency focus is going to expand to include the IT equipment itself, something that we think is overdue.”
Server heat on the rise
Data centres being built today need to remain economically competitive and technically capable for 10 to 15 years, however new chip technologies are causing operators to question traditional data centre design guidelines.
“Data-centre design must respond to server power and cooling requirements, and for years, these were consistent. Designers could plan for four to six kilowatts per rack,” said Daniel Bizo, research director at Uptime. “Successive IT refreshes did not require upgrades to power or cooling infrastructure.”
Now that’s changing. Power density per rack and per server chassis are escalating.
Intel’s 4th generation Xeon Scalable processors, code-named Sapphire Rapids, have a thermal design power (TDP) of up to 350 watts, for example, and AMD’s fourth-generation Epyc processors, code-named Genoa, have a TDP of up to 360 watts.
“Future product roadmaps are calling for mainstream server processors with 500- to 600-watt TDP just in the next few years,” Bizo said, “and so this trend is going to soon begin to destabilise facility design assumptions as we see mainstream servers approaching or exceeding one kilowatt each.”
Already, specialised high-performance computing (HPC) systems based on GPUs can require hundreds of watts per chip at peak power. In addition to high thermal power, they have lower temperature limits.
“They effectively place a double bind on cooling systems because they produce more thermal power, and many of them are also going to require lower operating temperatures,” Bizo said.
Removing a large amount of heat to reach a low temperatures is technically challenging, and that’s going to push operators to approach cooling differently, he said. For example, some data centre operators will consider support for direct liquid cooling.
The design dilemma posed by niche HPC applications can be considered an early warning of power-consumption and cooling challenges that high-TDP processors will bring to the mainstream enterprise server market.
“This now will require some speculation,” Bizo said. “What will be the power of a typical IT rack? How powerful will high-density racks become? What cooling modes will data centres need to support by the end of this decade?”
A conservative approach might be continuing with low-density rack designs, but that raises the risk of a data centre becoming too limited or even obsolete before its time.
However, a more aggressive design approach that calls for highly densified racks raises the risk of overspending on under utilised capacity and capabilities, Bizo warned.
“Operators are going to be faced with various choices in handling new generation IT technologies. They can restrict air temperatures and accept an efficiency penalty. Or as [US industry body] ASHRAE recommends with its Class H1 [thermal standard], they can create dedicated environments for these temperature-restricted IT systems to minimise that hit to facility efficiency," Bizo said.
"Or they can facilitate a conversion to direct liquid cooling.
“While semiconductor physics are the physical underlying driver here, infrastructure economics are going to shape how this turns out for data centre design and operations. And with little guidance from the chip makers, adaptability is going to be key.”
Energy efficiency focus to shift to IT
Pressure will continue to build for enterprises and data centre operators to deliver greater IT efficiency, Uptime predicts. In the past, the onus has been on the facility side.
Data centre operators, who bore the brunt of sustainability and efficiency demands, have made massive gains in power usage effectiveness (PUE) over the past 15 years or so, Bizo said. PUE measures the ratio of the energy used by the IT equipment to the energy used by the entire data centre.
Through initiatives such as air separation, more flexible temperature allowances, and tighter control of cooling, fans and power distribution, and use of renewable power sources, facilities are much more efficient in their resource use.
“Data centre operators have been doing quite a lot of things towards sustainability. At the same time, IT, by and large, has been absent from these discussions,” Bizo said. “This is becoming increasingly untenable. Why? Well, simply because, what’s left in terms of efficiency gains are hidden in IT energy performance.”
Pressure is building in four key areas, according to Uptime:
- Municipal resistance to new large data centres: Since 2019, concerns over power and land availability have led to greater restrictions on the construction of new data centres, and it’s likely to intensify. In Frankfurt, Germany, for example, new cloud and colocation builds are being restricted to specified zones out of concern for sprawling land use and changes to the city’s skyline.
- Limited availability of grid power to support increasing data centre capacity: Uptime cited two areas—Dublin, Ireland and northern Virginia—where the electrical grids are strained and power utilities have either paused or capped new connections because of shortfalls in generation or transmission capacity.
- Increasing regulation governing sustainability and carbon reduction, and more stringent reporting requirements: In the European Union, data centres are facing new, more detailed energy-consumption reporting requirements and will have to make their energy performance metrics publicly available, for example. Similar initiatives are now appearing in the US, according to Uptime.
- High energy costs: Energy prices and, consequently, power prices were on an upward trajectory before Russia’s invasion of Ukraine. Wholesale forward prices for electricity were already shooting up in both the European and US markets in 2021. The trajectory remains towards costlier power, and it’s creating an environment conducive to volatility.
Dealing with high energy prices, carbon reporting, and grid capacity shortfalls has always been the domain of facility operators.
But there’s diminishing returns on the facility side—greater interventions are delivering fewer and fewer benefits, according to Uptime. By contrast, IT is ripe to improve data centre energy performance, specifically in the area of server hardware.
Underutilised and inefficient server hardware is one key area where IT can improve energy performance. Having fewer servers with higher performance metrics could deliver energy gains, for example.
IT can make better hardware-configuration choices and use dynamic workload consolidation, Bizo notes. Power-management features such as energy-saving states and power throttling features, could also deliver major energy-efficiency gains.
The need to better architect IT infrastructure to deliver improved energy efficiency is only going to increase. Every watt saved by IT reduces pressures elsewhere, and reporting requirements will soon reveal the potential for greater energy efficiency hidden in IT infrastructure, Uptime asserts.
“IT infrastructure and operations teams will be expected, fairly soon, to demonstrate that they are doing everything they can to improve the typical utilisation of the server infrastructure,” Bizo said.
Geopolitics, cloud and data centre costs
To recap, two of Uptime’s 2023 predictions will see IT paying more attention to server hardware: power demands are rising as higher-powered server processors hit the market; and energy efficiency expectations are now landing on IT teams, not just facility teams.
Rounding out Uptime’s five data centre predictions for 2023 are:
Geopolitics will continue to deepen supply chain worries: Uptime cites political tensions between the US-led Western alliance, China and Russia that are raising the risk of supply chain disruptions.
Semiconductor supply chains and subsea cable systems are particularly vulnerable to potential economic and military confrontations, the firm predicts.
Cloud migrations will face greater scrutiny: In a period of economic strain and uncertainty, the cost of migration and the threat of spiralling cloud costs will slow down or deter some mission-critical migrations, Uptime predicts.
Data centre costs are set to rise: IT and data centre facility costs have been trending downward in recent years.
That trend has now ended, and prices are set to rise, Uptime predicts. Supply-chain problems, higher energy prices, and more expensive labor have all contributed to rising costs. That doesn’t mean higher prices are going to dampen demand for data centres, but they’re likely to drive renewed efforts to improve efficiency.