In the fall of 2024, researchers introduced an open-source tool called Perseus, which aims to reduce electricity consumption during the training of large language models like OpenAI’s GPT by up to 30%. This is achieved by optimizing the load on GPUs used for AI training through software.
Canadian computer scientists have discovered another simple way to significantly reduce the energy consumption of data centers. This involves optimizing the processing of network packets, as reported by the University of Waterloo. The current method of processing network packets in data centers is quite inefficient. A small change in the order of task execution could reduce power consumption by up to 30%.
Martin Karsten, a computer science professor at the University of Waterloo, compares this to optimizing a production process to avoid unnecessary movements. By restructuring the network packet processing process, Karsten and his team significantly improved CPU cache usage.
Only about 30 lines of Linux code need to be changed. Linux is used in data centers by all major tech companies. Companies like Amazon, Google, or Meta are cautious about changes. However, the researchers have already managed to get their modification included in the latest Linux kernel (6.13). Tech companies now just need to activate this change, which could save several gigawatt-hours of energy globally.
Almost every internet service could reduce the energy demand of individual requests, such as Google searches or ChatGPT prompts. The energy demand of server farms is expected to increase further in the coming years, especially due to AI.
According to a forecast by Goldman Sachs in May 2024, the “AI revolution” is expected to increase electricity consumption in data centers by 160% by 2030. Data centers worldwide are projected to require over 1,000 terawatt-hours of energy annually, compared to the current 400 terawatt-hours.
Efforts to optimize energy consumption in data centers are crucial, given the anticipated rise in demand driven by advancements in AI technologies. The adoption of efficient software solutions and strategic changes in data processing can contribute significantly to reducing the environmental impact of growing digital infrastructure.