At BitEnergy AI, developers have created an artificial intelligence that significantly reduces electricity consumption, possibly delaying the boom of nuclear power plants in the U.S.
In summary, they have managed to replace floating-point multiplication (FPM) with integer addition. The new method called "Linear Complexity Multiplication" (L-Mul) approaches the results of FPM while utilizing a simpler algorithm. It still maintains a high degree of accuracy compared to FPM results. This method can cut down energy consumption in AI systems by up to 95%, making it highly appealing for commercial applications.
Since this is a new process, existing accelerators from NVIDIA and others are not compatible with it—new hardware is required for practical use. The development of such equipment may necessitate significant investment, but the potential 95% reduction in energy consumption is likely to prompt major tech companies to take notice.
Google has shifted its climate goals to the back burner due to the energy demands of artificial intelligence: the company's greenhouse gas emissions have surged by 48% since 2019, instead of decreasing as expected. Google is now turning its attention to nuclear energy and other alternatives.
If the processing of artificial intelligence becomes energy-efficient, the benefits of AI technology could be realized without compromising the environment. Additionally, a 95% reduction in energy consumption would alleviate the pressure on national power generation systems and eliminate the need to construct numerous new power plants.
Sources: Tom's Hardware
Comments (0)
There are no comments for now