Presentation on theme: "Green Computing Energy in mobile games. Outline Introduction Reducing CPU power Reducing LCD power Acknowledgements: Slides partly by Ehsan AZIMZADEH."— Presentation transcript:
PID Controller-based Proportional–Integral–Derivative Generic closed loop controller Adjusting system parameters based on: – Difference between measured and set point K p : Proportional gain K i : Integral gain K d : Derivative gain Error = SP − PV
Experimental evaluation – Quake/Quake II game engines – Running on: Laptop – Intel Pentium M & Windows XP PDA – Intel XScale & Windows Mobile 5 Simu-disc – No transition overhead Simu- cont – No transition overhead and continuously scalable frequency
Energy Reduction Results Laptop – 20 fps Up to 22% less power than FIX Max possible saving: 23% – 30 fps 13% less power than FIX
Energy Reduction Results PDA – Under Simu-cont 35% less power than FIX – Under Simu-disc 25% less power than FIX – Upper bound on savings: 68%
LMS-based Low-Complexity Game Workload Prediction for DVFS B. Dietrich, S. Nunna, D. Goswami, S. Chakraborty, M. Gries TU Munich, Germany & Intel Labs, Germany ICCD 2010
LMS-based Workload Prediction Disadvantages of PID Controller-based: – Generality of PID Controllers – Unstability for most combinations of gains – Hand tuning – Different optimized values even for different runs of a game play Needs for dynamically adaptable predictor
Comparison: Linux Power Management Kernel ondemand governor – Sets V and F values for processor based on utilization – 7% power saving in this scenario – Why? LMS power saving – Up to 34.5%
Concluding Remarks DVFS effective in 3D games on mobiles Basic technique in these works: – Predict workload, set V and F accordingly PID controller – Unstable – Needs tuning several params LMS controller – Stable – Adjust only one param
Adaptive Display Power Management for Mobile Games National University of Singapore B. Anand, P. G Kannan J. Sebastian, A. L. Ananda M. C. Chan Singapore Management University K. Thirugnanam, R. K. Balan MobiSys’11, June 28–July 1, 2011, Bethesda, Maryland, USA.
Motivation Mobile games are becoming very popular. Display, wireless and CPU are The main power consuming elements on a mobile device. Power consumed by the LCD backlight is 45 to 95%.
Thin Film Transistor Liquid Crystal Display (TFT- LCD). There are three kinds of TFT LCD panels: – Transmissive – Reflective – Transflective Are used in: – computer monitors – mobile phones – handheld video game systems
Basic Idea The backlight component is the major source of power consumption. Brighten the frame Dimming Backlight Energy Savings
Increase the brightness Increase the pixel values Linearly: 1.Less quality. 2.Computationally expensive. Non-linearly
Non-linear approach use a standard Tone Mapping operator called the Inverse Gamma Function. Gamma Function widely available in: – OpenGL – X11 – game rendering engines – many commercial 3D games – mobile GPUs (Nvidia Tegra 2)
The Effect of Gamma and Linear Transformations. The Amount of Power Saved is the Same for Both Approaches.
Applying a higher Gamma value results in a higher contrast loss Global Contrast Loss vs Gamma
Different images will exhibit different quality loss Global Contrast Loss vs Image Brightness for γ=3
Backlight Power Measurement Laptop: Figure 12 setup HTC Magic & HTC Hero smartphones: using the Power Tutor
Gamma to Backlight Relationship set the Gamma and backlight intensity to their default values (1 Gamma and 255 backlight) The backlight levels ranged from 0 to 255 Gamma parameter ranged from 1 to 10
Computing Image Brightness Naïve approach – Compute average brightness of all images Careful sampling – 2000 samples (1 out of every 20 pixels)
Human Calibration of Gamma Thresholds Study the acceptable values of Gamma for each image brightness level. used 14 discrete brightness levels. 2 different images for each brightness level. The goal was to obtain an aggressive & conservative threshold for each image.
تمرین اضافی توان مصرفی گوشی موبایل خود و اجزای آن را با نرم افزار مناسب اندازه گیری کرده و تحلیل کنید. نوع گوشی، روش اندازه گیری، نوع صفحه نمایش و حالتهای مختلف آزمایش را گزارش کنید. بررسی کنید آیا گوشی شما با افزایش Brightness تصویر انرژی بیشتری مصرف می کند یا کمتری، و چرا؟