Greening computing infrastructure imperative

Editor's note: Accelerating the transition to green and low-carbon computing infrastructure is crucial for supporting the rapid development of artificial intelligence. China Environment News spoke to Su Ming, a researcher at the Energy Research Institute of the National Development and Reform Commission, on the development of green computing. Below are excerpts of the interview. The views don't necessarily represent those of China Daily.

The biggest challenge to developing green computing is matching computing loads with renewable power generation. Wind and solar power are inherently volatile, while AI inference workloads tend to fluctuate due to unpredictable user demand.

The second challenge is economic. Computing facilities are energy-intensive, and participating in green power trading or purchasing green power certificates entails a "green premium", which increases the cost of electricity to some extent.

Facilities opting for direct green power supply or integrated source-grid-load-storage models must invest in substantial new energy storage capacities. Additionally, maintaining operational safety requires a certain level of grid-connected capacity, which may not reduce the cost of using green electricity safely and reliably.

To achieve a green and low-carbon transformation of computing facilities, several strategic measures are necessary. First, emissions should be reduced at source by strengthening coordination between computing capacity and electricity supply. Computing facilities should maximize their use of renewable electricity. They should optimize their green energy supply models based on local conditions, participate in green power trading and purchase green power certificates.

New business models, such as direct green power supply, green microgrids and integrated source-grid-load-storage systems, should be encouraged. Additionally, exploring surplus cooling resources in surrounding areas can help meet the cooling demand of computing facilities.

Second, emissions should be reduced during operations by setting higher access standards and upgrading existing facilities. This aims to continuously lower the average power usage effectiveness, a core indicator of the energy efficiency of data centers that is calculated as the total facility energy divided by the IT equipment energy.

It is vital to deploy efficient cooling and heat-dissipation technologies, including liquid cooling, evaporative cooling, heat pipes and refrigerant pumps.

Enhancing the utilization of waste heat from computing facilities is also important, with facilities encouraged to develop heat-recovery systems for applications such as industrial-park heating, urban heating and agricultural facilities. Their intelligent operation and maintenance capabilities should be improved, alongside smarter management of energy consumption.

Finally, technological innovation must be strengthened to reduce the power consumption of computing itself.

AI models should be continuously optimized to improve computational efficiency, meeting the demands of inference, data storage and retrieval with lower power consumption rather than simply scaling up the number of computing cards, storage devices or model parameters. This will fundamentally improve energy efficiency and reduce power costs.