The Evolution of Data Centres: Traditional vs. AI-Optimised

In this rapidly evolving world of technology, data centres serve as the backbone of modern digital infrastructure. Traditionally, data centres have been the cornerstone of business operations, ensuring data storage, processing, and security. However, with the advent of artificial intelligence (AI), a new breed of data centre is required, designed to meet the specific demands of AI workloads and machine learning (ML) processes while also catered for the masses of cloud compute. Here, we explore the key differences between traditional data centres and AI-optimised data centres, highlighting the unique benefits each offer, with a focus on building services design including cooling requirements, and rack layout considerations.

Traditional Data Centres

Traditional data centres are engineered to handle a variety of general computing tasks, from managing web servers to storing large amounts of data. These facilities are designed with a focus on reliability, redundancy and availability, they are scalability, ensuring that businesses can maintain continuous operations and protect against data loss. Benefits include:

Reliability and Redundancy: Traditional data centres are built primarily with multiple layers of redundancy in power, cooling, and networking. This ensures high availability and minimises the risk of downtime, which is crucial for businesses that rely on constant access to their data.

Scalability: These data centres are designed to catering for scaleable rack densities and quantities, allowing businesses to add more servers and storage as their data needs grow. This scalability makes them suitable for a wide range of applications and industries.

Security: Traditional data centres prioritise physical and cyber security, implementing robust measures to protect sensitive hardware and software data from threats. This includes access controls, firewalls, and regular security audits.

AI-Optimised Data Centres

AI data centres are specialised facilities optimised to support the high-performance computing (HPC) requirements of AI and machine learning workloads. These data centres use cutting edge advanced hardware and software technologies to maximise efficiency and performance for their applications. Fundamental requirements include:

High-Performance Computing: AI and ML workloads require immense computational power, often involving complex algorithms and large datasets. AI-optimised data centres utilise specialised hardware such as GPUs (Graphics Processing Units) and TPUs (Tensor Processing Units) to accelerate these computations, delivering superior performance compared to traditional data centres. The workloads are shared between capable GPU’s meaning the data rack power profile is dynamic, a rack drawing 10kW may rise to 30 or 40kW over a short period of time putting strain on the electrical and mechanical systems. These systems need to be designed to cater for dynamic load profiles with quick response to match demand a mandatory requirement.

Rapid Deployment and Flexibility: AI data centres offer flexibility in deploying and scaling AI applications. With the integration of AI-specific infrastructure, these data centres can quickly adapt to the changing needs of AI research and development, allowing for faster innovation cycles.

Enhanced Data Management: AI data centres often include sophisticated data management tools that streamline the processing and analysis of large datasets. This facilitates faster data ingestion, preparation, and training of AI models, ultimately leading to quicker insights and decision-making.

Cooling Requirements and Techniques: Data centres generate significant heat, requiring effective cooling strategies to maintain optimal operating conditions. The cooling requirements for traditional and AI-optimised data centres are completely different in that traditional data centre loads are generally fixed within a known electrical load profile and any changes are planned whereas the AI-optimised data centre load is dynamic dependent upon workload and can vary in power requirements significantly due to the varying nature of their workloads and hardware. Next, we examine the different types of cooling systems traditionally used in data centres:

On-Chip Cooling: This method involves cooling directly at the chip level using microfluidic channels or other techniques to dissipate heat efficiently. It’s particularly relevant for high-performance computing in AI data centres but this solution cannot work alone as other elements of the AI process which are not chip cooled require cooling and therefore this system needs to be married to a second form of cooling system which is typically air cooled.

Immersive Cooling: Involves submerging hardware in a thermally conductive but electrically insulating liquid. This method provides excellent heat dissipation and is increasingly used in AI-optimised data centres due to its efficiency and reduced need for traditional air conditioning systems. Non immersive racks and system hardware will generally require some form of alternative cooling together with the immersive system. The immersive solution is the current gold standard in terms of effectiveness for dynamic loads and efficiency of cooling.

Air Cooling: Traditional data centres often rely on air cooling, where cooled air is circulated around servers to absorb and remove heat. This method is less efficient for high-density AI workloads but remains common in traditional setups.

Direct Liquid Cooling: Uses liquid coolants to transfer heat away from hardware components directly. This technique is highly effective for both traditional and AI-optimised data centres, particularly those with high-density configurations and substantial heat output.

The obvious crux is marrying the AI cooling with the alternative cooling solution to make efficient use of mechanical and electrical plant. Determining suitable water temperatures while optimising free cooling where feasible while catering for racks whose loading changes significantly within seconds. Reactive cooling solutions are too slow opening and closing vales to get the cooling medium to the racks and therefore proactive solutions need to be carefully considered within the context of energy efficiency and optimisation. The future is likely to look like a hybrid traditional data centre with AI and ML integrated into rack rows and the building services that support these also need to be hybrid in design.

Rack Layout and Scalability

The layout and scalability of racks within data centres are crucial factors in their design and operational efficiency.

Traditional Data Centres: Typically utilise a standardised rack layout with sufficient space for airflow and cooling. This setup is flexible and can be easily extended horizontally by adding more racks and servers as needed within predefined boundaries.

AI-Optimised Data Centres: Often require denser rack configurations to accommodate the high-performance hardware. The rack layout is designed to support advanced cooling techniques and facilitate rapid deployment of additional resources. This approach allows for vertical scaling, making it easier to integrate new technologies and expand capacity without significant redesign.

Both traditional and AI-optimised data centres play crucial roles in today’s digital landscape, each catering to different needs and applications. Traditional data centres offer reliability, scalability, and robust security, making them ideal for general computing tasks and business operations. On the other hand, AI-optimised data centres provide the high-performance computing power, energy efficiency, and flexibility necessary to drive advancements in AI and machine learning.

At BSE|3D, we understand the diverse needs of businesses and are committed to designing and implementing data centre solutions that meet your specific requirements. Whether you are looking to optimise your existing infrastructure or build a cutting-edge AI data centre, our team of experts is here to help you navigate the complexities and unlock the full potential of your data.

Contact us today to learn more about how we can support your data centre needs and help you stay ahead in the digital age. Email: mail@bse3d.com