Computations are cycled on the order of 3 trillion times per second in current computer processors. It's critical to keep track of the power used by such rapid transitions to preserve the chip's overall performance and efficiency. A processor can overheat and cause harm if it draws too much power. Internal electromagnetic problems may also occur because of sudden changes in power consumption, slowing down the entire processor.
Computer scientists have created a new AI system that can precisely estimate the power consumption of any computer processor at a pace of more than a trillion times per second while utilizing minimal processing effort. Because the technique has been proven on real-world, high-performance microprocessors, it is likely to enhance the Microprocessor and GPU Market. It could also improve the efficiency and inform the development of new microprocessors.
Creating the technique was based on the idea that software is needed to restrict undesirable extremes from occurring. This would also help computer engineers protect their hardware while enhancing their performance. All the present techniques are expensive and cannot keep pace with modern processors. Thus, the team decided to make an approach applicable to all types of microprocessors while also being feasible.
APOLLO (the technology) is based on an ideal power estimation algorithm that is precise and fast. Further, it can be easily integrated into a processing core while consuming little power. It could also become a standard component in future chip design because it can be utilized in any processing unit.
Artificial intelligence is the key to APOLLO's success. The proposed method uses artificial intelligence to identify and choose only 100 signals among a processor's millions that are most closely connected to its power consumption. It then takes those 100 signals and creates a power consumption model that it continuously monitors to predict the chip's performance.
This learning process can be implemented on almost any computer processor architecture, including those that have yet to be built because it is autonomous and data-driven. While the algorithm does not require any human designer's skills to function, it may assist human designers in their work.
After the AI has chosen its 100 signals, the algorithm can be examined to determine what they are. Many of the choices are straightforward, but even if they aren't, they can give designers feedback by revealing which operations are most strongly linked to power consumption and performance.
The method is unique and has the potential to be widely used. This is because it runs in the background on the microprocessor, which offers up a slew of new possibilities.