measuring per frame energy consumption of real time
play

Measuring Per-Frame Energy Consumption of Real-Time Graphics - PowerPoint PPT Presentation

Feb 17, 2023 •145 likes •889 views

Measuring Per-Frame Energy Consumption of Real-Time Graphics Applications Bjrn Johnsson & Tomas Akenine-Mller Lund University & Intel Corporation 1 Why care about energy? 2 Why care about energy? Computing devices

  1. Measuring Per-Frame Energy Consumption of Real-Time Graphics Applications Björn Johnsson & Tomas Akenine-Möller � Lund University & Intel Corporation � 1

  2. Why care about energy? � 2

  3. Why care about energy? Computing devices increasingly battery powered Visual & Parallel Computing Group � 3

  4. Why care about energy? Computing devices increasingly battery powered • Mobile phones • Laptops • Tablets • Wearable technology � Visual & Parallel Computing Group � 4

  5. Why care about energy? Computing devices increasingly battery powered • Mobile phones • Laptops • Tablets • Wearable technology Smaller form factors require graphics Visual & Parallel Computing Group � 5

  6. Why care about energy? Computing devices increasingly battery powered • Mobile phones • Laptops • Tablets • Wearable technology Smaller form factors require graphics • Heat dissipation • Battery size and time Visual & Parallel Computing Group � 6

  7. Why measure the energy? � 7

  8. Why measure the energy? Why not simulate? Visual & Parallel Computing Group � 8

  9. Why measure the energy? Why not simulate? • Simulation models need to be trained Visual & Parallel Computing Group � 9

  10. Why measure the energy? Why not simulate? • Simulation models need to be trained • Hardware speci fl c Visual & Parallel Computing Group � 10

  11. Why measure the energy? Why not simulate? • Simulation models need to be trained • Hardware speci fl c • Requires additional data: rendering time, primitive count… Visual & Parallel Computing Group � 11

  12. Why measure the energy? Why not simulate? • Simulation models need to be trained • Hardware speci fl c • Requires additional data: rendering time, primitive count… • Need to validate Visual & Parallel Computing Group � 12

  13. Why measure per-frame energy? � 13

  14. Why measure per-frame energy? Correlation Visual & Parallel Computing Group � 14

  15. Why measure per-frame energy? Correlation • Visually inspect energy-hungry frames Visual & Parallel Computing Group � 15

  16. Why measure per-frame energy? Correlation • Visually inspect energy-hungry frames • Other data collected per frame – Rendering time – Primitive count Visual & Parallel Computing Group � 16

  17. Energy � 17

  18. Energy Voltage U (volt) Voltage Time Visual & Parallel Computing Group � 18

  19. Energy Current I (ampere) Voltage Current Time Visual & Parallel Computing Group � 19

  20. Energy Power P (watt) � P = U*I Power Voltage Current Time Visual & Parallel Computing Group � 20

  21. Energy Energy E (joules) � ∫ E = P(t) dt Power Energy Time Visual & Parallel Computing Group � 21

  22. Our method � 22

  23. Our method Goal: Per-frame energy Visual & Parallel Computing Group � 23

  24. Our method Goal: Per-frame energy Method overview Three main steps Visual & Parallel Computing Group � 24

  25. Our method Goal: Per-frame energy Method overview Three main steps Voltage & current • Record current and voltage Time Visual & Parallel Computing Group � 25

  26. Our method Goal: Per-frame energy Method overview Three main steps P = U*I • Record current and voltage – To get power Time Visual & Parallel Computing Group � 26

  27. Our method Goal: Per-frame energy Method overview Three main steps • Record current and voltage Power – To get power • Acquire timestamps Time Visual & Parallel Computing Group � 27

  28. Our method Goal: Per-frame energy Method overview Three main steps • Record current and voltage Power – To get power • Acquire timestamps • Integrate power to energy Time Visual & Parallel Computing Group � 28

  29. Our method Goal: Per-frame energy Method overview Three main steps • Record current and voltage Power – To get power Frame 1 Frame 2 • Acquire timestamps • Integrate power to energy Time – For each frame Visual & Parallel Computing Group � 29

  30. Our method Step 1: Record current and voltage Recording • High frequency Data Acquisition Device (DAQ) Visual & Parallel Computing Group � 30

  31. Our method iPhone 4S Step 1: Record current and voltage Recording Battery • High frequency DAQ DAQ • Point of connection - + – What is measured? – What is not measured? Visual & Parallel Computing Group � 31

  32. Our method Step 1: Record current and voltage Recording • High frequency DAQ PSU • Point of connection CPU+ – What is measured? GPU – What is not measured? Motherboard DAQ iPhone 4S Battery DAQ - + Visual & Parallel Computing Group � 32

  33. Our method Step 1: Record current and voltage Recording DAQ • High frequency DAQ • Point of connection – What is measured? ATX +12V PSU – What is not measured? CPU+ GPU iPhone 4S PSU Motherboard CPU+ GPU Battery DAQ Motherboard - DAQ + Visual & Parallel Computing Group � 33

  34. Our method Step 1: Record current and voltage Recording • High frequency DAQ • Point of connection – What is measured? – What is not measured? DAQ iPhone 4S PSU CPU+ ATX +12V GPU PSU Battery CPU+ DAQ Motherboard GPU - DAQ + Motherboard Visual & Parallel Computing Group � 34

  35. Our method Step 2: Acquire timestamps � Intel Iris Pro Visual & Parallel Computing Group � 35

  36. Our method Step 2: Acquire timestamps Method 1: Recording • Altering measured workload – Requires source code • Synchronization required Intel Iris Pro Visual & Parallel Computing Group � 36

  37. Our method Step 2: Acquire timestamps Method 1: Recording • Altering measured workload – Requires source code • Synchronization required Method 2: Detect frame starts • Detect in curve – Semi-automatic – Works without source code Intel Iris Pro Visual & Parallel Computing Group � 37

  38. Our method Detecting frame starts � Intel Iris Pro Intel Iris Pro Intel Iris Pro Visual & Parallel Computing Group � 38

  39. Our method Detecting frame starts • Visually detect and mark � Mark Intel Iris Pro Intel Iris Pro Intel Iris Pro Visual & Parallel Computing Group � 39

  40. Our method Detecting frame starts • Visually detect and mark • Record surrounding � Record Intel Iris Pro Intel Iris Pro Intel Iris Pro Visual & Parallel Computing Group � 40

  41. Our method Detecting frame starts • Visually detect and mark • Record surrounding • Search for similar areas using LSE � Found Intel Iris Pro Intel Iris Pro Intel Iris Pro Visual & Parallel Computing Group � 41

  42. Our method Detecting frame starts • Where does the frame start? � � � Intel Iris Pro Intel Iris Pro Intel Iris Pro Visual & Parallel Computing Group � 42

  43. Our method Detecting frame starts • Where does the frame start? � � � Intel Iris Pro Intel Iris Pro Intel Iris Pro Visual & Parallel Computing Group � 43

  44. Our method Step 3: Integrate energy Integrate between timestamps end � ∑ P t t=start E = f Intel Iris Pro Visual & Parallel Computing Group � 44

  45. Our method Step 3: Integrate energy Integrate between timestamps end � ∑ P t ∫ t=start E = ≈ P(t) dt f Intel Iris Pro Visual & Parallel Computing Group � 45

  46. Observations � 46

  47. Observations Application launch • First few seconds of rendering iPhone 4S Visual & Parallel Computing Group � 47

  48. Observations Application launch • First few seconds of rendering • Signi fl cantly higher power iPhone 4S Visual & Parallel Computing Group � 48

  49. Observations Application launch • First few seconds of rendering • Signi fl cantly higher power – 5-10% higher – Both load and idle iPhone 4S Visual & Parallel Computing Group � 49

  50. Observations glFinish() GPU CPU Time Visual & Parallel Computing Group � 50

  51. Observations glFinish() • Blocks the application until commands are fl nished GPU CPU Time glFinish() Visual & Parallel Computing Group � 51

  52. Observations glFinish() • Blocks the application until commands are fl nished • Raises the power iPhone 4S iPhone 4S With glFinish() Without glFinish() Visual & Parallel Computing Group � 52

  53. Observations Disturbance by Operating System iPhone 4S Visual & Parallel Computing Group � 53

  54. Observations Disturbance by Operating System • Few frames with higher power iPhone 4S Visual & Parallel Computing Group � 54

  55. Observations Disturbance by Operating System • Few frames with higher power • 90ms every 5s iPhone 4S Visual & Parallel Computing Group � 55

  56. Observations Disturbance by Operating System • Few frames with higher power • 90ms every 5s – Same pattern without app running iPhone 4S Visual & Parallel Computing Group � 56

  57. Observations Disturbance by Operating System • Few frames with higher power • 90ms every 5s – Same pattern without app running Similar disturbances on most 
 platforms iPhone 4S Visual & Parallel Computing Group � 57

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Modelling the Energy Consumption of Soft Real-Time Tasks on Heterogeneous Computing Architectures

Modelling the Energy Consumption of Soft Real-Time Tasks on Heterogeneous Computing Architectures

Modelling the Energy Consumption of Soft Real-Time Tasks on Heterogeneous Computing Architectures H.E. Zahaf 1 , R. Olejnik 1 , G. Lipari 1 , A.E. Benyamina 2 1 Universit de Lille 2 University of Oran January 19, 2016 Outline Introduction

590 views • 36 slides
1 Original Frame Rate Original Frame Rate Instantaneous Frame Rate Instantaneous Frame Rate 60

1 Original Frame Rate Original Frame Rate Instantaneous Frame Rate Instantaneous Frame Rate 60

Real- -Time Rendering Time Rendering Real- Real -Time Rendering Time Rendering Real Performance Analysis (Echtzeitgraphik ( Echtzeitgraphik) ) Performance Analysis and Characterization and Characterization Dr. Michael Wimmer Dr.

360 views • 12 slides
Specialising Guavas Cache to Reduce Energy Consumption Nathan Burles 1 , Edward Bowles 1 ,

Specialising Guavas Cache to Reduce Energy Consumption Nathan Burles 1 , Edward Bowles 1 ,

Introduction Implementation Measuring Energy Consumption Experimentation and Results Conclusions References Specialising Guavas Cache to Reduce Energy Consumption Nathan Burles 1 , Edward Bowles 1 , Bobby R. Bruce 2 , Komsan Srivisut 1 1

654 views • 22 slides
Social and Real-time Web Applications using Meteor Developing Real-time Web Apps in JavaScript on

Social and Real-time Web Applications using Meteor Developing Real-time Web Apps in JavaScript on

Social and Real-time Web Applications using Meteor Developing Real-time Web Apps in JavaScript on Linux Anurag Bhandari A real-time application (RTA) is an application program that functions within a time frame that the user senses as immediate

688 views • 15 slides
Object-Oriented Genetic Improvement for Improved Energy Consumption in Google Guava Nathan Burles

Object-Oriented Genetic Improvement for Improved Energy Consumption in Google Guava Nathan Burles

Introduction Implementation Measuring Energy Consumption Experiments and Results Conclusions References Object-Oriented Genetic Improvement for Improved Energy Consumption in Google Guava Nathan Burles 1 , Edward Bowles 1 , Alexander E. I.

433 views • 19 slides
Real Time Human Islet Potency Assay Based on Oxygen Consumption Rate Klearchos Papas, PhD

Real Time Human Islet Potency Assay Based on Oxygen Consumption Rate Klearchos Papas, PhD

Real Time Human Islet Potency Assay Based on Oxygen Consumption Rate Klearchos Papas, PhD University of Minnesota Medical School Oxygen Consumption Rate Assessment Parameters Parameter Proportional To Measure of OCR Number of viable cells

460 views • 10 slides
RETIS Lab Real-Time Systems Laboratory Real-Time Scheduling for Multiprocessor Systems Giuseppe

RETIS Lab Real-Time Systems Laboratory Real-Time Scheduling for Multiprocessor Systems Giuseppe

RETIS Lab Real-Time Systems Laboratory Real-Time Scheduling for Multiprocessor Systems Giuseppe Lipari The multi-core revolution Technology trend use multiple processing elements on the same chip less power consumption, less

425 views • 6 slides
Energy Consumption in Energy Consumption in IP Networks IP Networks Rodney S. Tucker, Jayant

Energy Consumption in Energy Consumption in IP Networks IP Networks Rodney S. Tucker, Jayant

Energy Consumption in Energy Consumption in IP Networks IP Networks Rodney S. Tucker, Jayant Baliga, Robert Ayre, Kerry Hinton, Wayne V. Sorin ARC Special Research Centre for Ultra-Broadband Information Networks (CUBIN) University of

695 views • 21 slides
Today Intro to real-time scheduling Cyclic executives Scheduling tables Frames

Today Intro to real-time scheduling Cyclic executives Scheduling tables Frames

Today Intro to real-time scheduling Cyclic executives Scheduling tables Frames Frame size constraints Generating schedules Non-independent tasks Pros and cons Real-Time Systems The correctness of a real-time

441 views • 27 slides
Meaning of Energy Audit: Energy audit is an official scientific study of energy consumption of an

Meaning of Energy Audit: Energy audit is an official scientific study of energy consumption of an

Meaning of Energy Audit: Energy audit is an official scientific study of energy consumption of an organisation/process/plant/equipment aimed at reduction of energy consumption and energy costs without affecting productivity and comforts and

351 views • 12 slides
Kevin Fox ESCoE COVID-19 ECONOMIC MEASUREMENT WEBINARS UNSW Business School Centre for Applied

Kevin Fox ESCoE COVID-19 ECONOMIC MEASUREMENT WEBINARS UNSW Business School Centre for Applied

Measuring Real Consumption and CPI Bias under Lockdown Conditions Kevin Fox ESCoE COVID-19 ECONOMIC MEASUREMENT WEBINARS UNSW Business School Centre for Applied Economic Research Measuring Real Consumption and CPI Bias under Lockdown

778 views • 28 slides
Trading Off Lifetime, Fault-tolerance, and Power Consumption in Real-time MPSoC Jacopo Panerati

Trading Off Lifetime, Fault-tolerance, and Power Consumption in Real-time MPSoC Jacopo Panerati

Trading Off Lifetime, Fault-tolerance, and Power Consumption in Real-time MPSoC Jacopo Panerati , Samar Abdi , and Giovanni Beltrame eal, Concordia University Ecole Polytechnique de Montr MPSoC 2015 - Ventura Beach, CA,

332 views • 10 slides
How it works and what you need to know Clem son Energy Goals Reduce energy consumption 20%

How it works and what you need to know Clem son Energy Goals Reduce energy consumption 20%

How it works and what you need to know Clem son Energy Goals Reduce energy consumption 20% relative to the fiscal year of 2000 2020 10% increase of energy from renewable energy sources 2025 Promote sustainable energy projects Encourage

315 views • 13 slides
Measuring OS Support for Real-time CORBA ORBs Christopher D. Gill Washington University, St.

Measuring OS Support for Real-time CORBA ORBs Christopher D. Gill Washington University, St.

Measuring OS Support for Real-time CORBA ORBs Christopher D. Gill Washington University, St. Louis cdgill@cs.wustl.edu 27 January 1999 http://www.cs.wustl.edu/ cdgill/words-99. f pdf, ps g .gz Sponsors: Boeing, CDI/GDIS, DARPA contract

376 views • 14 slides
Promising Approaches Promising Approaches To Improved Infant d f Health Time Frame Time Frame

Promising Approaches Promising Approaches To Improved Infant d f Health Time Frame Time Frame

Promising Approaches Promising Approaches To Improved Infant d f Health Time Frame Time Frame 2011 Environmental Scan 2012 Commissioned Report From Preconception to Infant Protection Ph 1 G t 10 $681 000 Phase 1

291 views • 16 slides
Evaluating the trade-off between Performance and Energy Consumption in DAS-4 Performance and

Evaluating the trade-off between Performance and Energy Consumption in DAS-4 Performance and

Evaluating the trade-off between Performance and Energy Consumption in DAS-4 Performance and Energy Consumption in DAS-4 System and Networking Engineering Renato Fontana | Katerina Mparmpopoulou Presentation Flow Green concepts Project

473 views • 22 slides
Digital Microfilm Frame Detection Christopher Nelson Heath Nielson & Shane Hathaway The

Digital Microfilm Frame Detection Christopher Nelson Heath Nielson & Shane Hathaway The

Digital Microfilm Frame Detection Christopher Nelson Heath Nielson & Shane Hathaway The Church of Jesus Christ of Latter Day Saints Microfilm Frame Detection Scanning microfilm is much like taking pictures: 1. Scan a small strip of

289 views • 13 slides
CWE PRESENTATION RUBRIC Candidates Name: Presentation Topic:

CWE PRESENTATION RUBRIC Candidates Name: Presentation Topic:

CWE PRESENTATION RUBRIC Candidates Name: Presentation Topic: Date: Criteria:

157 views • 4 slides
Landmarks Preservation Commission Presentation 123 suLLivan street New York, NY, 10012 summarY:

Landmarks Preservation Commission Presentation 123 suLLivan street New York, NY, 10012 summarY:

Landmarks Preservation Commission Presentation 123 suLLivan street New York, NY, 10012 summarY: p.1 Cover page Courtyard terrace p.7 p.2 Site location Roof terrace p.8 p.3 Sullivan Street elevations Visibility from street p.9 p.4

359 views • 12 slides
Graph Methods for Multilingual FrameNets Collin F . Baker Michael J. Ellsworth International

Graph Methods for Multilingual FrameNets Collin F . Baker Michael J. Ellsworth International

The FrameNet lexical database as a set of graphs FN annotation Sentences Conclusions References Graph Methods for Multilingual FrameNets Collin F . Baker Michael J. Ellsworth International Computer Science Institute Berkeley, California

326 views • 28 slides
Extracting Information from Video Nichole Burgett Emily Ericson Background No way to get

Extracting Information from Video Nichole Burgett Emily Ericson Background No way to get

Extracting Information from Video Nichole Burgett Emily Ericson Background No way to get information out of videos currently Research is being done on algorithms for scene change detection Parallel algorithms written to process

322 views • 30 slides
CORPRATE Presentation MISSION STATEMENT Our mission is to provide quality solar products and

CORPRATE Presentation MISSION STATEMENT Our mission is to provide quality solar products and

CORPRATE Presentation MISSION STATEMENT Our mission is to provide quality solar products and services in the most efficient and responsible manner, generating outstanding customer value, while remaining committed to the principles of

673 views • 21 slides
PERFORMANCE IN VDI WITH NVIDIA GRID JASON SOUTHERN SENIOR SOLUTIONS ARCHITECT FOR NVIDIA GRID

PERFORMANCE IN VDI WITH NVIDIA GRID JASON SOUTHERN SENIOR SOLUTIONS ARCHITECT FOR NVIDIA GRID

PLANNING FOR DENSITY AND PERFORMANCE IN VDI WITH NVIDIA GRID JASON SOUTHERN SENIOR SOLUTIONS ARCHITECT FOR NVIDIA GRID AGENDA Recap on how vGPU works Planning for Performance - Design considerations - Benchmarking Optimizing for Density

562 views • 41 slides
ZEN AND THE ART OF VGPU SELECTION Jeremy Main - Lead Solution Architect NVIDIA GRID, Japan

ZEN AND THE ART OF VGPU SELECTION Jeremy Main - Lead Solution Architect NVIDIA GRID, Japan

ZEN AND THE ART OF VGPU SELECTION Jeremy Main - Lead Solution Architect NVIDIA GRID, Japan jmain@nvidia.com The real purpose of the scientific method is to make sure nature hasnt misled you into thinking you know something you actually

1.07k views • 90 slides

More recommend