BitEnergy AI developers were able to reduce the energy consumption of AI computing systems by 95%

The increase in energy consumption associated with the development of artificial intelligence is becoming one of the most urgent problems in the technological industry. The need for computing power, necessary for training and operation of AI models, is rapidly increasing. Already today, powerful speech models such as ChatGPT consume a significant amount of energy every day – about 564 MW∙h, which is enough to power about 18 thousand American homes. Forecasts for the coming years suggest that by 2027 data center (DTC) energy consumption for AI could reach 85 to 134 TWh per year.

To solve this problem, some companies are considering the use of small nuclear reactors that can power AI-powered data centers due to their stable and highly efficient energy production. Instead, other developers are looking for new approaches to optimize computing to reduce power consumption.

BitEnergy AI: A New Approach to Energy Saving

One of the promising solutions is the offer of BitEnergy AI, which has developed a new Linear-Complexity Multiplication algorithm. It allows you to significantly reduce the power consumption during computational operations by replacing traditional floating-point calculations with integer operations. This is especially important because floating point operations are the most energy intensive in AI data processing.

• Linear-Complexity Multiplication involves the use of approximation methods that replace complex operations with simpler integer additions. This approach allows you to save up to 95% of energy for tensor multiplications and about 80% for scalar products.
• Testing has shown that the new algorithm is capable of maintaining the required accuracy, even exceeding current 8-bit computing standards, and performing better on popular models such as Llama and Mistral.

Problems and prospects of implementation

The main difficulty is that the implementation of the new algorithm will require the development of special hardware, since existing chips such as Nvidia do not support this approach. However, according to the company, there are already plans to create specialized hardware and software APIs that will be able to take advantage of the new method.

This innovation could be an important step towards reducing the operating costs of AI and reducing the carbon footprint, which is in line with the global trend towards sustainable development and environmentally friendly technologies.

It is noteworthy that according to a report by The Wall Street Journal, Apple abandoned plans to invest in OpenAI, a surprise twist ahead of the close of OpenAI’s $6.5 billion funding round. Microsoft and Nvidia remain potential investors, with Microsoft already owning 49% of OpenAI’s profits and planning to invest another $1 billion.

Apple’s decision to end the financing talks came as a surprise given that it had previously been reported that the company might participate in the investment. However, Apple rarely invests in large Silicon Valley technology companies, the sources said. Despite the disinvestment, Apple and OpenAI will continue their collaboration, which will allow ChatGPT to be integrated into iOS 18 later this year. This partnership will allow ChatGPT to work together with Siri to handle requests.

Interestingly, neither party will generate any financial payout from this partnership, which Apple believes will bring it value comparable to or greater than the cash investment.