
Researchers Find Ways For AI
Processing To Save Energy
Jakarta, CNBC Indonesia –The implementation of artificial intelligence (AI) technology is currently a threat to environmental sustainability. This is because computers or hardware used to run AI are considered wasteful of electricity or not energy efficient.
Now, a number of scientists from BitEnergy, a company with a mission to design environmentally friendly AI models, have created an algorithm so that AI can run and be implemented optimally and remain energy efficient.
This AI algorithm, called Linear-Complexity Multiplication (L-Mul), is claimed to be able to reduce energy consumption from AI processing by up to 95 percent.
BitEnergy explains in detail how the L-Mul algorithm can save power in the following document. Simply put, the algorithm will insert a simpler mathematical calculation method in AI processing.
Just like computer processing, AI processing, in one process, also involves many mathematical numbers or binary numbers called floating-point numbers.
These numbers are used by AI to carry out their tasks in predicting or solving something. Well, the L-Mul algorithm here functions to make the floating-point number calculation process simpler by adding several mathematical formulas.
Ultimately, AI processing will be more energy efficient, because it can be done efficiently.
In BitEnergy’s trial, L-Mul is claimed to not only save power, but is also said to be able to minimize performance degradation in AI processing. Usually, energy savings do have one downside, namely decreased performance.
They also claim that L-Mul can be integrated into several popular AI models such as GPT, Llama, Mistral, and so on. In several trials of a number of these AI models, BitEnergy claims that L-Mul can also increase the accuracy of AI in carrying out its tasks.
Although it has a positive effect, BitEnergy said that L-Mul cannot be implemented just like that. This is because this algorithm requires special hardware in order to run according to its function.
This is reasonable because solutions that involve computing processes like AI usually also have to be accompanied by technology or hardware updates. The good news is, BitEnergy plans to develop the hardware, along with an application system (API) that can embed the L-Mul algorithm into existing AI models, as summarized by KompasTekno from TechSpot, Thursday (10/10/2024)
BitEnergy did not mention when the hardware and API will be available and ready to be adopted.
When it is available, the trend of adopting energy-saving AI through L-Mul will of course also depend on adoption by large companies that run AI such as Google, Microsoft, OpenAI, and so on.
For example, technology giant Google announced in its environmental report that the company’s greenhouse gas (GHG) emissions have jumped 48 percent in the past five years.
The main cause is the skyrocketing electricity consumption of data centers to support its new artificial intelligence (AI) products.
The company also revealed that emissions in 2023 had increased 13 percent compared to the previous year, reaching 14.3 million metric tons
SOURCE : CNBC INDONESIA