Can innovation solve the PUE challenge and enhance DC sustainability?
Even before the exponential rise of AI computing power, and the corresponding rise in rack power density, data centres have already been among the most energy and water intensive infrastructure classes. Each facility, on average, uses around 40% of their power on cooling alone.
According to consultants ADP, a large data centre can consume the equivalent energy to 50,000 homes. Cooling the rows and racks of high performance computers also requires up to 19 million litres of water a day – a similar quantity of water to that consumed by a small city of 50,000 people.
Data centre operators know only too well they must innovate to be energy-efficient, particularly if they are going to hit 2030 climate goals. And energy consumption has been complicated by energy costs rising as much as 16% [a median figure] according to Cushman and Wakefield.
Government action has led DC owners to act
Last November, the Federal Government announced new data centres selling services to the government will need to achieve at least a five-star National Australian Built Environment Rating System (NABERS) energy rating by mid-2025 – based on their actual operational emissions – so Australia, in this respect at least, wants to get ahead of the game. NextDC has already hit the five-star rating at its Sydney (S1) and Melbourne (M1) facilities, for example.
However, for DC operators and builders, even the location of a DC will influence its energy use and this provides opportunities for off-site renewable energy PPAs to help operators hit their emission reduction targets and reduce costs. Equinix recently signed a renewable PPA with TagEnergy totalling 151MW. Once operational in 2029, the PPA will supply enough energy to the grid to cover Equinix’s seventeen IBX data centres across Australia. NTT last week did the same for its largest DC in Japan.
Climate change will also create new challenges for Australia and New Zealand DC operators who will need to watch Singapore closely on how to deal with hot climates. Last year the island state launched the world’s first standard for optimising energy efficiency for data centres in tropical climates.
Is the industry doing enough on sustainability?
The W.Media Melbourne convention later this month [21st March 2024] will be exploring AI, Automation and Future Operations and several panels and keynotes will debate whether DCs owners and builders have the tools needed to meet the demand arising from these trends while running sustainable operations.
The DC industry also needs to explain itself better to the wider world or risk being branded serial polluters/”energy guzzlers”.. A UTS report, prepared for Pure Storage last year, found that almost one in three sustainability professionals in Australian companies did not even consider data centre energy consumption – despite 81% of respondents suggesting their demand for data management would be increasing. Even more damning was that only 5% of respondents felt that the quality of sustainability-related data received from data centre operators was detailed. 46% of respondents were receiving no sustainability-related data at all from their DC partners.
Reducing PUE through innovation
Many in the industry understand the need to better articulate how the sector can be sustainable. To lower power usage effectiveness (PUE), data centre operators are already deploying new software solutions and algorithms to gain efficiencies. But higher rack density exacerbates the cooling demands. In 2022, on average, DC PUEs were near 1.55 and now the push is to get the PUE down towards 1.00.
As a result, liquid cooling is now an inevitability given it deals with rising power consumption but when should DC operators take the plunge? It’s already becoming cost-competitive with conventional solutions but questions need to be answered such as: what types of hardware are compatible with immersion cooling? Are there any limitations or special requirements? How does immersion cooling affect the lifespan and performance of hardware components such as CPUs, GPUs, memory modules, and storage devices? What types of cooling fluids are commonly used in immersion cooling systems, and what are the environmental implications of each option? How easily can an immersion cooling system be integrated into an existing data centre infrastructure?
The innovation is happening already. Tasmanian-based company Firmus has become a leader in highly scaled immersion cooled computing platforms, tying up with ST Telemedia Global Data Centres to launch GPU-centric Infrastructure as a Service offering, focused on deep learning AI and visual computing workloads. This is called Sustainable Metal Cloud and it is pushing the boundaries of PUE close to 1.00 despite the Nvidia H100 GPUs and Dell HPC kit it is packing.
And in Korea, SK Telecom just signed a deal with UK-based precision liquid cooling specialist Iceotope Technologies to develop a liquid cooling technique/solution that optimises the energy-efficiency of AI data centres.
Using heat as power
Other companies have been looking at using waste heat from data centres to generate power for neighbours. Data centres are a constant heat output source so technologies like thermoelectric generators or Organic Rankine Cycle (ORC) could allow the DC operator to harness waste heat.
In a white paper by Vertiv looking at an ORC system, for example, replicating a 30kW liquid-cooled high performance compute rack of servers in data centre application has been designed, fabricated, and tested to capture the server waste heat and convert a portion to electricity while transporting the heat to ambient air. The full load power usage effectiveness (PUE) was 1.02 compared to the 1.6 to 2.2 range in small data centres.
Beyond PUE for true sustainability
While PUE is a valuable metric for assessing energy efficiency in data centres it has to be used with caution when assessing a facility’s energy performance. For example, some data centre operators and enterprises may selectively report data to paint a more favourable picture of their energy efficiency. They might exclude certain energy-consuming components or areas from their calculations to artificially lower their PUE.
PUE calculations also often assume full load conditions, which may not accurately reflect real-world scenarios. In addition to this, many operators will be referring to design PUE rather than operational PUE – the actual energy performance during day-to-day operations.
Therefore, the time has come to debate some form of overall resource efficiency measure for data centres that would go beyond energy use and factor in things like water usage, materials usage, carbon emissions, and waste management within the data centre.
To be truly sustainable, innovation shouldn’t just target energy efficiency but should also address reducing water usage, minimising waste generation, and adopting environmentally friendly materials. This echoes what Scope 3 emission reduction measures try to do in supply chains. Given the growing concerns about water scarcity, e-waste, and carbon emissions, it presents another chance for innovation. Such a measure would require unprecedented industry collaboration to standardise metrics and methodologies. However, to be a truly sustainable sector, those discussions need to happen.
[Author: Simon Dux]
Comments