Are you ready for AI?
Data centres are in the midst of a seismic shift; already heavily invested in the energy transition towards renewables, Artificial Intelligence (AI) is now firmly accelerating the need for that transition while simultaneously creating new demands and opportunities of its own.
AI is moving rapidly and is power hungry. AI is demanding; it creates heavy & spiky workloads that necessitate augmented IT infrastructures with even greater stability and reliability than ever before. AI impacts everyone it touches, increasing complexity for the operator and adding pressure to the grid.
With annual data creation reaching unsurpassed and perhaps previously unimagined levels, by loading AI into the mix, the explosion in the demand for data is tangible. GOLDMAN SACHS forecasts that data centre electricity consumption could drive 160% increase in data centre power demand. – representing more than 30% of total Data Centre power consumption - further strengthening the case for greater efficiencies and a focus on renewable energy sources.
Our dependency on data is real and growing which, in turn, impacts every aspect of a data centre and its operation – from sheer computing power to supporting infrastructure, from design specifications to the nature and scale of the wider facility, both in terms of planning, construction and ongoing support.
“We’re already aware of increasing challenges in handling HPC (High Performance Compute) load profiles due to their fluctuating nature. Rapid load cycles, voltage instability and the long-term impact in equipment and battery lifespan are being flagged as concerns by operators already coming up against some of the complexities that AI can present.”
Given the pace with which AI is growing and the immense amount of power that a proliferation of synthetic data and applications will consume, how can you ensure that your equipment and services needed to support those facilities will be AI-ready?
Clement BARTHELMEBS continues; “It’s important to think about AI-readiness in the context of some of the big headlines – without losing focus of the fundamentals. For instance, ensuring that your UPS is tailored to supply and react appropriately to the load profile – while meeting exacting standards and staying compliant when it comes to dynamic performance.
Underpinning this, it’s vital to ensure that your UPS is extremely reliable with a high MTBF and design rules that enable the handling of unusual loads. Secondly, it’s vital to consider the level of protection required for the distribution, ensuring compatibility of the UPS even under worst case operating scenarios, when working on the Genset or during network losses. Advanced Genset management is also critical in terms of delivering robust solutions that are fit for AI.
Finally, harnessing the power of predictive maintenance for enhanced reliability is going to prove invaluable as the creation of synthetic data ramps up.”
Socomec’s range of UPS is well known to provide very high reliability based on very high MTBF and our stringent design rules make Socomec UPS’ able to handle this kind of non-usual customer load.
The emergence of AI and its unique load behaviour
With AI evolving at full throttle, it’s of paramount importance that power supply systems have the capability to perform reliably and to remain stable in the face of a growing demand for compute-intensive tasks.
AI applications often exhibit unique load characteristics, such as high gradient load fluctuations from 0 to more than 100%. This means that it is crucial to have an Uninterruptible Power Supply (UPS) that can handle these demands seamlessly if a facility is to fulfil its potential. Without robust UPS solutions, there is a risk that the gains from AI are lost if power is interrupted.
Clement explains; “AI systems often involve complex computations that require significant and instantaneous bursts of power. For supplying power for deep machine learning models or HPC (high performance computing) can suddenly spike power consumption, going from zero to full load in a fraction of a second. This unique load behaviour—characterised by large load steps and high variability—demands a UPS that can respond with precision and speed.”
Traditional UPS systems, designed for more predictable and steady loads, may struggle to keep up with the dynamic power requirements of AI applications. Additionally, these sudden and significant power demands can pose serious challenges to the electrical grid. The grid must be able to handle these peak power loads without compromising stability.
“When multiple AI systems simultaneously demand high power, it can lead to grid instability, including voltage fluctuations and frequency variations. This not only affects the AI systems themselves but can also have broader implications for other critical infrastructures reliant on the same power supply.”
Socomec DELPHYS XL tailored for AI applications
Powering everything from data analytics and machine learning to autonomous systems and sophisticated robotics, the rapid growth in AI applications has brought about new challenges in the realm of power management.
In response to these challenges, Socomec has developed a groundbreaking high-power UPS that excels in the management of AI loads – ensuring uninterrupted performance and optimal efficiency.
Engineered specifically to handle rapid fluctuations, the system ensures that AI systems receive a consistent and reliable power supply. By smartly managing energy consumption, Socomec’s DELPHYS XL reduces stress on the network, helping to maintain grid stability while protecting AI applications.
Meeting stringent standards in terms of dynamic performance: IEC 62040-3 compliant.
Socomec’s DELPHYS XL high power UPS has undergone rigorous testing and validation by independent third-party laboratories. These tests confirm that the UPS meets the dynamic performance requirements set by IEC 62040-3, achieving Class 1 compliance, the most stringent classification within this standard. This compliance guarantees that the UPS can handle large load steps without causing any disturbance to critical loads, maintaining stability and reliability even under the most challenging conditions.
The impact of AI on power sources
Managing sensitive equipment in parallel with inputs such as network stability, customer demands require smart coordination between UPS real time consumption and energy storage capacity during idle periods to support the grid. This new UPS functionality will be crucial with the rise of AI, as it introduces the challenge of balancing battery capacity to assist the grid multiple times a day while ensuring long-term battery health despite frequent cycling. Power utility instability has long been a challenge for data center operators, but it is now even more critical with the emergence of AI and its uniquely demanding load patterns.
This becomes especially apparent when equipment is powered by a generator, for example, as stability and reliability are vital in order to manage the impact of repetitive loads and to mitigate against voltage and frequency drops. Such instabilities can result in costly interruptions with the potential to cause damage to hardware as well as corrupting data, culminating in many lost computational hours and wasted energy.
Clement BARTHELMEBS explains; “When a Socomec UPS is supplied by the mains, advanced features come into play in order to enhance grid stability. A range of settings smooth the input consumption, matching it to the maximum capability of the grid and taking into account the acceptable gradient of power. This helps prevent the network from experiencing significant power peaks, which can lead to instability or stress on the grid also opening doors to new power sources that are more sensitive to load variation such as fuel cells or renewable energy.
“Another advanced capability is peak shaving, where the UPS manages energy consumption by setting the maximum power acceptable by the network. By doing so, it reduces the overall stress on the electrical grid and enhances the reliability of the power supply.
“In scenarios where the main power supply is unavailable, however, and the UPS operates on a generator, maintaining the stability of the generator becomes critical. Generators are particularly sensitive to repetitive load impacts, which can lead to dips in voltage and frequency, ultimately causing disruptions for the connected loads.”
Socomec advanced genset management for AI applications
Socomec's UPS dynamically manages generator stability by continuously monitoring its frequency, preventing detrimental effects, and ensuring uninterrupted power continuity.
Clement explains; “Here's how it works: when the generator experiences instability and its frequency and voltage begin to decline, the UPS intervenes intelligently by temporarily shedding part of the load. The energy is smartly managed between the generator and the batteries for a brief duration, ensuring uninterrupted power supply to critical loads. This proactive load management assists the generator in regaining its nominal frequency, thereby ensuring a stable power supply without interruptions.
“This advanced genset management capability is the result of Socomec's extensive experience in energy storage applications and grid support services for UPS systems. Over the years, Socomec has developed sophisticated technologies and solutions to enhance power stability and reliability, which have now been integrated into our UPS products, providing superior generator support and load management that’s ready for AI applications.”
