As the pace of digital transformation continues, a number of Fortune 500 companies are looking increasingly at Artificial Intelligence (AI) applications to support their future business growth. Suitable for use across a broad range of industries, AI can assist with optimisation, preventative maintenance and chatbots together with fraud and anomaly detection.

Its success relies on the availability of vast amounts of data and has therefore resulted in an increased demand for scalable data centre services that have the capacity to store and process this huge volume of information.

Historically, the demand for data centres arose purely from an IT storage and computing requirement. More recent years have seen the migration away from data being held on-premises to cloud infrastructure, the advancements in new software applications and Internet of Things (IoT) technology prompting significant growth in this area and also therefore in the level of data centre inventory. 

AI machine learning however, has different requirements altogether. To meet the demand data centres are having to evolve once more, adapting their design, power infrastructure and cooling capacities to support the vast computing resources and storage requirements that new AI applications and workloads demand.

At the moment, there are still a number of unknowns on how AI will fully affect data centre development and demand, but one thing is for certain ….. we are moving progressively towards a world that embraces AI and whether stored in-house or via a third party, we need to be prepared.

 

Let’s Take a Closer Look at AI

In its simplest term, AI is the science of building a machine so that it thinks like a human. Unlike a human however, AI technology can recognise patterns, inform decisions and make judgements on enormous amounts of complex data in record time.

Each AI application requires an immense amount of computing power to support its two machine learning functions:

• AI training – building a model from the input of a huge dataset
• AI inference – generating predictions, solutions and actionable results from dataset learnings

Each function has its own unique storage and power needs. Data is supported through high-performance computing (HPC) clusters, that consist of multiple servers connected through high-speed networks. This set-up allows for parallel processing and faster training times.

 

High-Performance Computing (HPC)

A HPC system will often be designed to fit into a standard 19 inch wide four-post rack – a common configuration for data centre equipment that is capable of housing rack-mounted servers, blade servers, networking equipment and storage arrays. Modular and scalable, the rack system allows for easy capacity upgrades as needs change.

The power density within a single rack can range from 20 kW to over 60 kW therefore the appropriate power, cooling and back-up protection infrastructure must be in place to ensure the hardware is able to function as it should.

 

HPC Hardware

HPC Systems use a combination of high-powered processors (CPU’s), high-speed memory and specialised hardware such as Graphics Processing Units (GPU’s) or Tensor Processing Units (TPU’s).

These processors are able to perform complex analyses swiftly and accurately, allowing the HPC system to handle a multitude of operations, such as; scientific simulations, data mining, advanced analytics and machine learning tasks.

 

Data Storage

The growth of data generation fuelled by AI is transforming how this information is now stored, processed, managed and transferred whilst simultaneously increasing the demand for computing power across both cloud and edge data centres.

Aside from the extensive storage requirements, AI applications perform numerous input and output functions; for example exchanging information between devices across a communications network and/or reading data or writing it to varying storage devices. High-speed storage and access is therefore essential for AI workloads that demand the swift retrieval of data to produce ‘real-time’ results. 

 

Networking

The extensive calculations produced through AI workloads demand high-capacity, error-free networks. As AI becomes more highly adopted this will ultimately place greater strain on the data centre in terms of bandwidth and capacity.   

Large-scale AI applications generally require approximately three times more bandwidth than traditional computing networks. This requirement becomes greater still where the varying components of an AI application are distributed across a number of hardware and software assets on different servers and storage systems.

 

Power

The high-performance processers that AI requires draw more power than those of traditional data centre processors, increasing the overall power usage and power density within each facility.

This demand presents its own unique energy efficiency and sustainability challenges. As the processing requirements generate a significantly higher level of heat within the space, more efficient cooling technologies are needed to reduce the potential for equipment failure and downtime. This may involve a level of re-engineering or redesign to implement new liquid or immersion cooling techniques that will ensure the appropriate temperatures can be maintained.

    

What Types of Data Centre can be used for AI?

As high power density racks are the most suitable for AI applications and workloads, they can be utilised across a number of different sized facilities; from small edge data centres all the way through to large hyperscalers.

The smaller, edge data centres are typically housed in locations close to where the data is being generated and used. They are therefore more suitable for low-latency AI applications that require real-time access, such as; virtual reality applications, video analytics, autonomous vehicles, drones and augmented reality interactive experiences.

Larger hyperscale data centres, on the other hand, are more suited to AI workloads that involve machine learning (ML), large volumes of data analytics, deep learning (DL) training, natural language processing (NLP) and computer vision. 

 

AI within the Data Centre Environment

Data centre operators are already utilising forms of AI in their daily operations as a way to boost performance. Examples of this include monitoring the facility’s hardware to detect and fix equipment issues, improve energy efficiency, check temperature and humidity levels, and also facilitate operation-based technologies.  

Similarly, it can be used to support a number of customer-oriented services such as chatbots, onboarding processes, fraud detection, data visualisation and marketing analytics.  

Ensuring these services are continually maintained demands a constant power supply and therefore, also a reliable and robust back-up power protection strategy. Already a pre-requisite for any data centre environment, emergency power will remain at the forefront of all design schemes intended to support future AI applications. Adopters will need to scale their strategies accordingly to safeguard against the catastrophic consequences that will arise from loss of such considerable volumes of client data.

Whilst we, mere humans, may not yet be fully aware of how AI will affect our lives, like it or not, its on the horizon. We are already adopting artificial intelligence in a number of formats but its set to get much bigger and smarter. What is important is that we look ahead towards our design needs for the future and start considering the solutions now that offer long-term scalability within our data centre environments.   

A leading provider of UPS systems that support data centre needs, Power Control has been supplying, installing and maintaining power protection strategies for over 30 years. We work with a variety of UPS manufacturers to develop unique solutions that deliver reliability and resilience, whilst also taking into account both the current and long-term objectives of each business.

If you would like information on how Power Control can help to future-proof your data centre needs, please call a member of the team on 01246 431431 or email info@powercontrol.co.uk