Green Computing in Data Centers: Definitions, Research Scope, and Energy‑Saving Technologies

Background

With the rapid development of societal informationization, the reliance on information systems across government, industry, and society has dramatically increased, leading to high energy consumption in the information industry. Data centers alone account for about 1.8% of total societal electricity use, making energy efficiency a critical concern.

Definition of Green Computing

Green computing, though lacking a universally accepted definition, generally encompasses three layers: energy and resource conservation, efficient and circular utilization, and environmental friendliness (low carbon and non‑harmful). It involves using high‑efficiency, low‑power devices and infrastructure, rational resource allocation while ensuring service reliability, and achieving low‑cost, low‑energy sustainable systems.

Research Objects

Analysis of literature shows that roughly 70% of green‑computing research focuses on data centers, cloud computing, and servers, with the remainder covering CPUs, GPUs, scheduling, resource management, space utilization, and environmental impacts such as CO₂, noise, and radiation.

Applications and Energy‑Saving Effects in Data Centers

The article reviews energy‑saving technologies across four layers: IT systems, power distribution, air‑conditioning, and overall architecture.

1. Dynamic Voltage and Frequency Scaling (DVFS)

DVFS reduces chip power by lowering supply voltage and clock frequency in tandem, following a workflow that collects load signals, predicts required performance, translates performance to frequency, and adjusts voltage accordingly. Major processors (e.g., Intel SpeedStep/EIST, AMD PowerNow, Cool&Quiet) already support DVFS.

2. Heterogeneous Computing

Combining general‑purpose CPUs with specialized units such as GPUs or DSPs enables workload‑specific acceleration. CPU+GPU platforms are mature, while CPU+MIC, CPU+FPGA, and other heterogeneous solutions are also being explored by companies like Intel, IBM, Facebook, Microsoft, and Baidu.

3. Liquid‑Cooling Server Technology

Liquid cooling uses high‑specific‑heat fluids to remove heat, with three main implementations: immersion cooling, cold‑plate cooling, and spray cooling. It can reduce air‑conditioning capacity by up to two‑thirds and is adopted by vendors such as Sugon, Huawei, Lenovo, HP, Cisco, 3M, and Fujitsu.

4. Rack‑Level (Whole‑Cabinet) Server Technology

Integrating power and cooling at the cabinet level reduces fan count by over 93%, cuts cooling power by more than 25%, consolidates power modules to improve conversion efficiency from ~85% to ~94%, and doubles space utilization, with major Chinese vendors like Inspur leading development.

5. Direct Mains + UPS/HVDC Power Supply

Traditional UPS‑plus‑battery solutions achieve only 80‑90% efficiency due to redundancy. Direct mains combined with UPS or HVDC can match reliability while significantly improving efficiency, and emerging configurations (e.g., one‑line mains + one‑line UPS or HVDC) are becoming industry trends.

6. Backplane Air‑Conditioning

Backplane cooling places cooling coils directly behind each cabinet, providing close‑to‑source heat removal, eliminating hotspots, and saving floor space by removing separate room‑level air‑conditioners.

7. Fluorocarbon Pump Dual‑Cycle Air‑Conditioning

This smart dual‑cycle system uses outdoor ambient cooling in winter and compressor‑based cooling in summer, achieving an annual energy‑efficiency ratio (AEER) of 7.37 versus 3.91 for conventional units, saving roughly 46.9% of energy.

Conclusion

Guided by recent Chinese government policies on energy‑saving and green data‑center construction, green‑computing technologies are increasingly deployed in new data centers, achieving PUE values around 1.5. The trend points toward customized, modular, and virtualized large‑scale data centers with higher power density and lower PUE.