Can air-cooling handle the heat from a 1-GW data centre?, Pg11

Key Highlights:

• Google's planned 1-GW data center in Visakhapatnam will reportedly use air-cooling technology for its servers.
• Data centers generate significant heat, with a 1-GW facility producing 1 GW of heat that needs to be dissipated.
• Air-cooling is less water-intensive but faces limitations, especially with high-density modern AI chips like Nvidia Blackwell.
• Google has committed to replenishing 120% of water consumed for non-cooling needs by 2030, quantified under the Volumetric Water Benefit Accounting standard.
• Tata Consultancy Services (TCS) HyperVault, another 1-GW facility, plans to use liquid cooling, including Direct-to-Chip (DTC) cooling, and renewable energy.

Data Center.jpg

Detailed Insights:

• Heat in data centers is generated by billions of transistors in processors as they manipulate electrical currents.
• Air-cooling involves fans pushing cool air through server racks, with various configurations like CRACs, CRAHs, and free-cooling.
• Direct liquid cooling, such as cold-plate cooling, uses liquids with higher heat capacity than air for more efficient heat transfer.
• Immersion cooling involves submerging electronics in a non-conductive liquid, offering high power density handling but requiring specialized hardware.
• Evaporative cooling is energy-efficient in non-humid conditions but demands large quantities of water.
• Dry-cooling avoids water use but requires larger heat-exchange surfaces, especially in warm climates.
• Air-cooling generally has lower upfront infrastructure costs but can incur higher total ownership costs due to performance and efficiency limitations.
• The "thermal wall" for air-cooling is around 40 kW per rack, while modern AI chips can generate 120-150 kW per rack.
• Hybrid cooling, combining air-cooling for low-density areas and liquid cooling for high-density AI clusters, is a common approach for large data centers.
• Google's water management plans include rainwater catchment, groundwater recharge, and water ATMs.

Key Concepts Involved:

• Data Center: A facility housing computer systems and associated components, such as telecommunications and storage systems.
• Hyperscaler: A large-scale data center operated by major cloud providers, characterized by massive computing and storage capabilities.
• Heat Capacity: The amount of heat energy required to raise the temperature of a substance by a specific amount.
• Thermal Wall: The practical limit of heat dissipation that can be achieved using air-cooling technology in data centers.