AI will be responsible for power management in Intel Meteor
Hardware

AI will be responsible for power management in Intel Meteor Lake processors

Already in May it became known that Intel Meteor Lake processors will use a special hardware VPU engine, which will be responsible for certain tasks related to the operation of AI algorithms. However, Intel also plans to use AI to control the power consumption of systems based on these processors.

    Image source: Intel

Image source: Intel

The new AI-driven performance algorithm will be used in all of the company’s future products, Efraim Rotem, responsible for consumer SoC architecture at Intel Engineering Group, said during the Hot Chips conference at Stanford University. According to him, the new feature will appear in Meteor Lake processors, which are expected to be released within the next two months.

The graph below shows how the new Meteor Lake AI algorithm (highlighted in orange) manages CPU power consumption compared to traditional algorithms for different workloads. The result is an increase in the energy efficiency of the chip.

With the new AI algorithm, Intel also wants to increase the responsiveness of its processors. “We place special emphasis on the responsiveness of the system when interacting with a computer. We want immediate action and don’t want to wait too long.”Rotem noticed.

The most obvious solution to improve system responsiveness is to boost performance, which requires more power to be directed to the chip. This allows him to work faster and complete tasks faster. However, after a given task completes, the CPU needs to determine when that task completes and then go to sleep. This technology is called “Dynamic Voltage and Frequency Scaling” or DVFS.

“Power management is all about how we determine the right processor speed.”Red says.

Intel pioneered the core functionality of this technology in its 6th generation Core processors (Skylake). There it was called Speed ​​Shift. The feature was responsible for switching the processor from a high-power active state to a lower-power standby mode. However, its capabilities were limited to specific use cases. For example, it worked when opening and closing web browsers.

In Meteor Lake, the capabilities of the AI ​​algorithm to manage processor power will increase significantly. Now the algorithm “understands” and can predict when and how the user will open a web page, scroll through it, close it and open another. The same monitoring algorithm is used for many other tasks. The difference is that the AI ​​algorithm of the new processors is self-learned. And based on that, I extracted more detailed patterns of behavior than those previously programmed by Intel in the same Skylake.

Using an AI power management algorithm, Meteor Lake chips are up to 35% more responsive. That means they get into a state of heightened performance more quickly, Rotem says. At the same time, the algorithm enabled processors to go to sleep much faster when there was no load. This has made Meteor Lake up to 15% more energy efficient.

Basically, it’s about giving the processor the power budget it needs for the time it needs, and no more. According to Rotem, there is significant room for improvement in the technology. In its current form, the AI ​​algorithm is trained for specific scenarios. Therein lies the limitation. It is already trained and does not respond dynamically to individual user preferences. In other words, the computer no longer learns from the behavior of the user. At least not in this generation of AI algorithms. Rotem also suggested that different AI models could be applied to different scenarios. For example in games.

About the author

Dylan Harris

Dylan Harris is fascinated by tests and reviews of computer hardware.

Add Comment

Click here to post a comment