New integer addition algorithm can reduce AI power consumption by 95%

The new technique uses integer addition instead of floating point multiplication.
A team of engineers at BitEnergy AI, an AI inference technology company, has reported a method to reduce the power consumption of AI-based applications by 95%. The team published a paper describing their technology on the arXiv preprint server.
As AI applications have become mainstream, their use has skyrocketed, leading to a marked increase in energy requirements and costs. Large linguistic models such as ChatGPT require a lot of computational power, which in turn means that they require a lot of power to run. For example, ChatGPT now requires about 564 MWh daily, enough to power 18,000 American homes. As science continues to advance and such applications become more popular, critics have suggested that these applications could consume around 100 TWh annually in just a few years, on par with bitcoin mining operations.
The BitEnergy AI team claims that with the new development, they have managed to find a way to significantly reduce the amount of computation required for AI applications without resulting in performance degradation.
The new technique is quite nifty - instead of using complex floating point multiplication (FPM), the method uses integer addition. Applications use FPM to handle extremely large or small numbers, allowing applications to perform calculations using it with great precision. It's also the most energy-intensive part of AI number processing.
The researchers call their new method Linear-Complexity Multiplication - it works by approximating the FPM using integer addition. They claim that testing so far has shown that the new approach reduces power consumption by 95%.
The only drawback is that it requires different hardware than what is currently in use. But the research team also notes that the new type of equipment has already been designed, built and tested. However, it's still unclear how such hardware will be licensed - currently, the artificial intelligence hardware market is dominated by GPU maker Nvidia. How they react to this new technology could have a significant impact on the pace of adoption - if the company's claims are confirmed.