Home Innovation Data Centers Data Centers Snip Energy Consu...

Data Centers Snip Energy Consumption by 30% with a Few Lines of Codes


Data Centers

Data Centers Cut Energy Use by 30% with Optimized Code

According to recent studies, data centers may cut their energy use by as much as 30% by making minor changes to about 30 lines of code in the network stack of the Linux kernel.

Interrupt request suspension, a method that maximizes CPU power efficiency by minimizing needless disruptions during periods of high traffic, is the AI innovation. Generally, a new packet of data entering the network causes an interrupt request, which slows down operations by asking the CPU core to cease its present activity in order to process the data.

By enabling the system to actively scan the network for fresh data packets as necessary rather than waiting for each individual interrupt, the new code lowers the number of interrupt requests. However, when traffic slows down, the system switches back to interrupt handling because this method uses a lot of power.

The kernel's handling of IRQs can be improved to increase data throughput by up to 45% while maintaining low tail latency. In other words, for the most time-sensitive processes, the system can manage additional traffic without experiencing any delays. The latest version of the Linux kernel now includes the change.

By 2030, data centers will account for up to 4% of the world's power consumption, at least partially due to AI. With 1.76 trillion parameters, training OpenAI's GPT-4 used as much energy as 5,000 American homes' worth of electricity annually.


Business News


Recommended News

Latest Magazine