Explore chapters and articles related to this topic
Dynamic Voltage Scaling
Published in Vojin G. Oklobdzija, Digital Design and Fabrication, 2017
Although DVS was not even considered feasible in commercial products three or four years ago, the rapidly evolving processor industry has begun to adopt various forms of DVS In 1999, Intel introduced SpeedStep®*, which runs the processor at two different voltages and frequencies, depending upon whether the notebook computer is plugged into an AC outlet, or running off of its internal battery [22].In 2000, Transmeta introduced LongRun®*, which dynamically varies voltage and frequency over the range of 1.2–1.6 V and 500–700 MHz, providing a 1.8 times variation in processor energy consumption. Control of the voltage/frequency is in firmware, which monitors the amount of time the operating system is sleeping [23].In 2000, AMD introduced PowerNow!®*, which dynamically varies voltage and frequency over the range of 1.4–1.8 V and 200–500 MHz, providing a 1.7 times variation in processor energy consumption. Control of the voltage/frequency is implemented via a software driver that monitors the operating system's measure of CPU utilization [24].In 2001, Intel Inc. introduced the XScale®* processor, which is essentially the second generation StrongArm®* processor. It can dynamically operate over the voltage and frequency range of 0.7–1.75 V and 150–800 MHz, providing a 6.3 times variation in processor energy consumption, the most aggressive range announced to date. By further advancing the energy-efficiency of the original StrongArm, this device will be able to deliver 1000 MIPS with average power dissipation as low as 50 mW at 0.7 V, yielding an ETR as low as 0.05 μW/MIP2 [25].
Power-Aware Characteristics of Matrix Operations on Multicores
Published in Applied Artificial Intelligence, 2021
Guruprasad Konnurmath, Satyadhyan Chickerur
For power management, the techniques are broadly classified among two major categories: reactive and predictive (Khaled, El-Hosseini, and Ali 2015). The reactive techniques react in accordance to changes in performance based on workload. When the change is identified in workload’s state, accordingly this technique responds to that specific change. Hence, predictive technique is preferred to be better than reactive technique. Our research work is followed by predictive technique. To optimize power consumption in CPU as well GPU, the most well-known technique is DVFS method (Sparsh and Jeffrey 2014)(Ashish and Khare 2015)(Xinxin et al. 2013). This method is implemented by altering the voltage and frequency levels of multicores. This technique is introduced commercially under many names such as AMD’s PowerNow, Intel’s SpeedStep technology and many more.