• • • DATA CENTRE MANAGEMENT • • •
Inside Liquid Cooling:
Innovation in the Data Centre By David Watkins, solutions director at VIRTUS
rtificial Intelligence (AI) applications, spanning machine learning and deep learning algorithms, demand extensive computational power to process vast amounts of data and perform complex tasks. This computational intensity translates into significant heat generation within data centres – the infrastructure that underpins AI advancements and ensures their continuous operation. Efficient and effective cooling solutions are paramount to ensure optimal performance, reliability and longevity of IT systems, putting data centre operators under significant pressure to innovate and integrate advanced cooling technologies capable of efficiently handling the heat generated by AI applications.
A Advanced cooling technologies
for AI workloads Traditional air-cooled systems, commonly employed in data centres, may struggle to effectively dissipate the heat density associated with AI workloads. As AI applications continue to evolve and push the boundaries of computational capabilities, innovative cooling technologies are becoming indispensable. Liquid cooling methods offer efficient heat dissipation directly from critical components, mitigating the risk of performance degradation and hardware failures associated with overheating. Liquid cooling can also be a more sustainable option than other thermal management technologies. It can reduce the amount of energy used by a facility as it takes less electricity to cool a server than air cooling systems. It is important to understand that whilst liquid cooling supports far denser compute deployments, it is not detrimental to a facility’s Water Usage Effectiveness (WUE) performance.
There are several types of liquid cooling systems being adopted in data centres. Immersion cooling involves submerging specially designed IT hardware (servers and graphics processing units, GPUs) in
a dielectric fluid, such as mineral oil or synthetic coolant. The fluid absorbs heat directly from the components, providing efficient and direct cooling without the need for traditional air-cooled systems. This method significantly enhances energy efficiency and reduces the running costs, making it ideal for AI workloads that produce substantial heat. Direct-to-chip cooling, also known as microfluidic cooling, delivers coolant directly to the heat-generating components of servers, such as central processing units (CPUs) and GPUs. This targeted approach maximises thermal conductivity, efficiently dissipating heat at the source and improving overall performance and reliability. By directly cooling critical components, the
direct-to-chip method helps to ensure that AI applications operate optimally, minimising the risk of thermal throttling and hardware failures. This technology is essential for data centres managing high-density AI workloads.
Advantages of a hybrid cooling strategy
The versatility and flexibility of liquid cooling technologies provides data centre operators with the option of adopting a mix-and-match approach tailored to their specific infrastructure and AI workload requirements.
Integrating multiple cooling solutions enables providers to optimise efficiency and address varied cooling needs. Different types of liquid cooling can be deployed in the same data centre, or even the same hall. By combining immersion cooling, direct-to-chip cooling and / or air cooling, providers can leverage the benefits of each method to achieve optimal cooling efficiency across different components and workload types. As AI workloads evolve and data centre requirements change, a flexible cooling infrastructure that supports scalability and adaptability becomes essential. Integrating multiple
cooling technologies provides scalability options and facilitates future upgrades without compromising cooling performance. For example, air cooling can support High
Performance Computing (HPC) and AI workloads to a degree, and most AI deployments will continue to require supplementary air-cooled systems for networking infrastructure. All cooling types ultimately require waste heat to be removed or re-used, so it is important that the main heat rejection system (such as chillers) are sized appropriately and enabled for heat reuse where possible.
Challenges to adoption A hybrid cooling strategy, which harnesses liquid cooling innovation can be beneficial for efficient thermal management. However, with this innovation comes challenges to adoption, not least of which is the initial investment required for implementing the liquid cooling infrastructure. Overcoming this challenge requires careful cost-benefit analysis and long-term planning to demonstrate the return on investment (ROI) of liquid cooling in terms of energy savings and performance improvements for AI workloads.
Another challenge is the complexity of liquid cooling system design and integration. Unlike air-based cooling systems, liquid cooling solutions require specialised components, such as cooling distribution units (CDUs) which must be carefully integrated into existing data centre infrastructure. This means that retrofitting older data centres can be expensive as well as complex.
New data centres are likely to be better suited to support HPC and AI workloads because they have been built with these new demands in mind. Data centre providers should invest in skilled personnel and training to effectively design, deploy and maintain liquid cooling systems tailored to the unique requirements of AI workloads.
Integrating liquid cooling is the way forward
Effective cooling solutions are paramount if data centres are to meet the ever-growing demands of AI workloads. Liquid cooling technologies play a pivotal role in enhancing performance, increasing energy efficiency and improving the reliability of AI-centric operations.
The adoption of advanced liquid cooling technologies not only optimises heat management and reuse but also contributes to reducing environmental impact by enhancing energy efficiency and enabling the integration of renewable energy sources into data centre operations.
16 ELECTRICAL ENGINEERING • NOVEMBER 2024
electricalengineeringmagazine.co.uk
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50
