Google promises a smaller AI energy footprint, but has not yet reached the goal.

Just like any other electronics and services, AI also has its energy footprint. Google says that it is working day and night to reduce the AI's energy footprint to a level where it can be considered 'non-existent.' However, the way to reach the goal is not quite easy.

One Gemini query consumes about five drops of water

Google claims in its latest report that it was able to achieve a 30-fold reduction in the energy consumption levels of the Gemini AI over the last year. The company adds that the median energy consumption per text prompt has been reduced by a factor of 33x. This is over a period of 12 months. The associated carbon footprint has decreased by more than 44x.

Google says that it has been able to achieve this because of the "full stack" approach. This maximizes energy efficiency while ensuring optimal results. The company adds that this approach involves optimizing every single layer of the AI. Model architectures, such as Mixture-of-Experts (MoE), algorithms, and Tensor Processing Units (TPU), have to be redesigned in a way that consumes a lower level of energy and resources. Implementing this requires a lot of research and development, adds the brand.

One median text prompt of the AI consumes about 0.24 watt-hour (Wh) of energy. This is equivalent to the energy consumed by a TV for about nine seconds.

Google aims to reduce the AI's energy footprint to almost zero

Although the energy footprint of Gemini is lower than its competitors such as ChatGPT, the tech giant still wants the energy footprint to decrease to a level where it is essentially "negligible."

The Vice President of Engineering at Google, Amin Vahdat, says, "I don't think we're anywhere close to where we need to be. But we've gotten it to a place where it's comparable to other sources of information."

The postGoogle promises a smaller AI energy footprint, but has not yet reached the goal.appeared first onAndroid Headlines.