AWS announces large-scale datacenter reform for the next generation AI era, aiming to achieve up to 46% cooling efficiency improvement through the introduction of liquid cooling systems, strengthening support for high-density servers essential for AI processing. Simultaneously, significant enhancement in stability and efficiency is expected by simplifying the power supply system. These innovations are projected to increase the energy efficiency of data centers by up to 4.1 times compared to traditional methods.
AWS has announced a large-scale datacenter renovation for the next generation AI era. By introducing a liquid cooling system, it achieves a maximum 46% cooling efficiency improvement, enhances support for high-density servers essential for AI processing, and significantly improves stability and efficiency through the simplification of the power supply system. These innovations are expected to increase the energy efficiency of datacenters by up to 4.1 times compared to conventional methods.
Strengthening AI infrastructure through the introduction of a liquid cooling system.
AWS's new cooling system represents an ambitious technological innovation to cater to the next generation AI workloads. At the core of this system is the 'Chip Direct Liquid Cooling' method that involves direct contact liquid cooling with high-performance AI processors such as the in-house developed Trainium2 chip and NVIDIA's GB200 NVL72. This method allows for efficient heat control even in high-density computing environments where conventional air cooling systems struggled to adapt.
It is noteworthy that this new system is designed in such a way that it can be retrofitted into existing datacenters.This means that AWS's existing infrastructure can be upgraded step by step, minimizing downtime and costs associated with large-scale facility updates.
Furthermore, in the development of the liquid cooling system, AWS is closely collaborating with major chip manufacturers. NVIDIA's Vice President Ian Buck emphasizes the importance of this partnership by stating that it enables efficient cooling of AI infrastructure while minimizing energy consumption.
The notable technical feature of the system is its multimodal design that can flexibly combine liquid cooling and air cooling. Liquid cooling is applied to servers performing high-performance AI processing, while conventional air cooling is used for equipment with relatively low heat generation such as network devices and storage servers. This allows for the selection of the optimal cooling method based on the characteristics of each equipment.
The effectiveness of this new cooling system has already been demonstrated, confirming that mechanical energy consumption required for cooling during peak times can be reduced by up to 46%. Particularly notable is the fact that this significant efficiency improvement does not result in increased water usage. This can be considered an important achievement from an environmental impact perspective.
The notable technical feature of the system is its multimodal design that can flexibly combine liquid cooling and air cooling. Liquid cooling is applied to servers performing high-performance AI processing, while conventional air cooling is used for equipment with relatively low heat generation such as network devices and storage servers. This allows for the selection of the optimal cooling method based on the characteristics of each equipment.
The effectiveness of this new cooling system has already been demonstrated, confirming that mechanical energy consumption required for cooling during peak times can be reduced by up to 46%. Particularly notable is the fact that this significant efficiency improvement does not result in increased water usage. This can be considered an important achievement from an environmental impact perspective.
Fundamental review of the electrical supply system.
The new electrical supply system introduced by AWS represents an ambitious initiative to fundamentally review the reliability and efficiency of datacenters. The core of this reform lies in simplifying the path of electricity delivery from the transmission grid to the servers as much as possible. In conventional datacenters, electricity needed to undergo multiple conversions before reaching the servers from the transmission grid. This not only resulted in efficiency losses but also increased the risk points for system failures.
In the new system, the number of electricity conversions has been significantly reduced to minimize energy losses during conversions. Additionally, a design where backup power sources are placed closer to the server racks has been adopted. This has shortened the emergency power supply route, enhancing the system's reliability. This design change has successfully reduced potential failure points in the electrical system by as much as 20%.
In the new system, the number of electricity conversions has been significantly reduced to minimize energy losses during conversions. Additionally, a design where backup power sources are placed closer to the server racks has been adopted. This has shortened the emergency power supply route, enhancing the system's reliability. This design change has successfully reduced potential failure points in the electrical system by as much as 20%.
An innovative approach has been implemented in the heat dissipation system. While conventional systems used numerous electric fans to forcibly expel heat, the new system adopts a design that actively utilizes natural air pressure differences. This has enabled efficient heat dissipation while significantly reducing the number of fans used. This approach not only reduces fan power consumption but also decreases system complexity and contributes to improved maintainability.
As a culmination of these improvements, the newly developed power shelf is noteworthy. This innovative design allows the electrical density per rack to increase by a factor of 6 over the next 2 years, and further triple after that. This is not merely a numerical goal, but a strategic move to meet the extensive computational requirements of next-generation AI workloads.
As a culmination of these improvements, the newly developed power shelf is noteworthy. This innovative design allows the electrical density per rack to increase by a factor of 6 over the next 2 years, and further triple after that. This is not merely a numerical goal, but a strategic move to meet the extensive computational requirements of next-generation AI workloads.
Comprehensive approach to reducing environmental impact.
Prasad Kalayanaraman, Vice President of AWS Infrastructure Services, stated that these datacenter features are an important step in improving energy efficiency and adapting flexibly to new workloads. In fact, the introduction of this new system makes it possible to increase computing power per site by 12% and reduce the total number of required datacenters.
Efforts to reduce environmental impact go beyond improving cooling systems. Innovation extends to the construction materials of datacenters, successfully reducing carbon emissions in the concrete manufacturing process by 35% compared to the industry average. Additionally, renewable diesel is used in backup generators, leading to a 90% reduction in greenhouse gas emissions over the lifecycle compared to traditional fossil fuel-derived diesel.
Efforts to reduce environmental impact go beyond improving cooling systems. Innovation extends to the construction materials of datacenters, successfully reducing carbon emissions in the concrete manufacturing process by 35% compared to the industry average. Additionally, renewable diesel is used in backup generators, leading to a 90% reduction in greenhouse gas emissions over the lifecycle compared to traditional fossil fuel-derived diesel.
In response to the sharp increase in power consumption in datacenters due to the full-scale emergence of the AI boom, AWS's current initiative demonstrates a very fundamental approach. In particular, the adoption of a hybrid liquid and air cooling system can be considered a practical solution that balances efficiency and cost.
However, a cautious technical verification is necessary for the plan to significantly increase power density by 6 times in 2 years and then a further 3 times. Managing localized heat concentration associated with high density poses a potentially high-risk critical challenge that could impact system stability.
Another noteworthy aspect is that these technological innovations are designed considering retrofits for existing datacenters. This demonstrates that sustainability is not just a slogan but is incorporated as a practical design philosophy.
However, a cautious technical verification is necessary for the plan to significantly increase power density by 6 times in 2 years and then a further 3 times. Managing localized heat concentration associated with high density poses a potentially high-risk critical challenge that could impact system stability.
Another noteworthy aspect is that these technological innovations are designed considering retrofits for existing datacenters. This demonstrates that sustainability is not just a slogan but is incorporated as a practical design philosophy.
December 3, 2024
Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only.
Read more
Comment
Sign in to post a comment
