Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Thermal-Scheduling

Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Thermal-Scheduling

Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Thermal-Scheduling For Ultra Low Power Mobile Microprocessor George Cai1 Chee How Lim 1 W. Robert Daasch2 Intel Corporation 1 Integrated Circuit Design and Test Laboratory PSU 2 May, 2002 1 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Presentation Outline Mobile CPU Power Efficiency With Demanded Performance Thermal Scheduling For Mobile Microprocessor Power Constrained Performance Observations/Conclusions May, 2002 2 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Ultra Low Power Mobile Microprocessor Primary FE DE Secondary DE

OOP EX Majority mobile apps with performance requirements RF IOP Text email, caller-id, reminder and other none high performance w/ anywhere-anytime requested apps Primary pipeline: maximal performance, complex pipeline structure Second pipeline: Minimum power and energy consumption, very simple in order structure and target mobile anywhere-anytime applications. Transparent to OS and applications Maximal utilizing on die clock/power gating for energy saving May, 2002 3 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Low Energy Consumption With Providing Suitable Performance Is Key For Anywhere And Anytime Must be compatible with exist OS and platform Must have active leakage power control Must meet the real time telecom application requirements Stock/Urgent Messages All urgent message And important news News headline Email titles Calendar reminder Pages/voice message Alert Interactive command and reply

May, 2002 Stock Update 4 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Runtime Thermal Scheduling Capability Primary FE DE Secondary DE OOP EX RF IOP When thermal threshold is exceeded, the pipeline clusters will service instructions in alternating manner: cool the hot pipeline by clock/power-gating & the cold pipeline sustains processor operations Flexible selecting the threshold point, the energy-delay product, performance, and reliability of the processor can be enhanced May, 2002 5 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Thermal Effects: Leakage Trend Active leakage power reduction will be significant role for total power reduction Thermal control is important for low energy consumption for mobile CPU Leakage Power Trend Leakage/Total Power 0.6

~25% power reduction 0.5 0.4 0.25um 0.18um 0.13um 0.10um 0.3 0.2 0.1 0 30 50 70 90 110 T emperature (degree C) Derived from F. Pollacks Micro-32 Keynote Presentation, 1999 May, 2002 6 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Example of Scheduling Algorithm T1 TL & T2 TL T1 > TL || T2 > TL Tmax TH T1 TL S4 TS1 Temperature (C)

tcycle tcool theat T1 > TL & T2 TH TS2 T1 < TH T1 TH S2 T1 TL TL Ta S1 T1 TH S3 T1 > TL & T2 < TH Time (s) S1: Normal Operation (Primary Pipeline) S2: Stall Fetch & Clear Pipeline S3: Alternate Operation (Secondary Pipeline) S4: Disable Clock or Scale F-V May, 2002 7 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Enhance Effectiveness Of Other Power Control Techniques Dynamic Frequency Scaling PLL DCT LOGIC

RENAME ARCH FILE EXE I L 1 FETCH ROB LSB DL1 DECODE TS TS BYPASS DECODE I L 1 FETCH TS PLL Global Clock May, 2002 ROB LSB RENAME ARCH FILE EXE

DL1 Global Clock DFS LOGIC VRM TS BYPASS Dynamic Clock Disabling/Throttling On-Chip Vdd Off-Chip 8 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Power Constrained Clock Frequency With Performance Impact Using gate delay as proxy for performance C Vdd Vdd td I ds D Vdd Vt Thermal-dependence: mobility & threshold Relative Operating Frequency 1.20 1.00 0.80 D 300 Normalized 0.60 F ref 300 T

2 Vt Vt 300 2.5 mV K T 300 Frequency impact (B subscript denotes baseline) 0.40 30 15 0.60 0.500 0.80 0.70 1.00 Fraction Vdd 0.90 90 75 60 0.00 45 0.20 Temperature (C) Fref Vdd Vt Vdd ,B D F

FB D , B Vdd , B Vt B Vdd May, 2002 9 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Thermal Effects on Power Divide total power into two components: dynamic and leakage power 2 P C Vdd F Vdd I s e Vt nVT Relative Power Consumption 1.80 1.60 1.40 1.20 Normalized 1.00 Pref 0.80 0.60 30 15 0.500 0.60 0.80 0.70 1.00 Fraction Vdd 0.90

75 60 0.00 45 0.20 May, 2002 90 0.40 Temperature (C) 10 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Thermal Effects on Energy Using power per frequency (W/MHz) metric as proxy for energy Energy Power Frequency Relative Energy Consumption 2.50 2.00 1.50 Normalized Eref 1.00 30 15 0.500 0.60 0.80

0.70 May, 2002 90 75 60 1.00 Fraction Vdd 0.90 0.00 45 0.50 Temperature (C) 11 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Architecture-Level Power-Performance Tradeoff For wide-superscalar processors, performance impact of pipeline scaling is smaller than global clock throttling or frequency scaling Relative Frequency vs Supply Voltage Relative IPC vs Pipeline Width 1.40 1.4 1.30 Relative IPC ~15% 1.10 1.00 0.90

0.80 0.70 Relative Frequency 1.2 1.20 1.0 ~30% 0.8 0.6 0.4 0.2 0.60 0.0 0.50 '2 '4 '8 '16 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 Fraction of Supply Voltage Pipeline Width May, 2002 12 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Comparative Outcomes: Energy Metric Simulation Conditions (500 million instructions; TL = 55C) Stop Clock Control: Toggle between Fmax and 0 MHz

Voltage/Freq Scaling : Thermal Scheduling : Toggle between Fmax and 0.9/0.8/0.6 Fmax Toggle between Primary and 2nd Pipelines Conservative: TH = 70C Aggressive: TH = 60C Energy Consumption of Aggressive Control Energy Consumption of Conservative Control 14.000 10.000 8.000 Benchmarks 6.000 4.000 PERL GCC 2.000 LI 0.000 Clk gating V-F scaling M88KSIM Thermal ControlThermal scheduling Techniques M88KSIM Energy (J) Energy (J) 12.000 16.000 14.000 12.000 10.000 8.000

6.000 4.000 2.000 0.000 Benchmarks PERL GCC LI Clk Gating LI GCC M88KSIM F-V Scaling Thermal Scheduling Thermal Control Techniques PERL May, 2002 M88KSIM LI GCC PERL 13 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Comparative Outcomes: Energy-CPU Time Metric Total Energy x CPU Time Overall Energy-CPU Time Metric Comparison 16.00 Energy-Time (J.s) 14.00 12.00 10.00 Overall_Cv (J.s) 8.00

Overall_Ag (J.s) 6.00 4.00 2.00 0.00 DCD DFS TAM Thermal Control Techniques May, 2002 14 Thermal-Scheduling For Ultra Low Power Mobile Microprocessor Pros and Cons Advantages Limits power/energy upper bound & prevents thermal runaway Pipeline tuned for either performance or ultra low power Existing OS and application compatible Performance penalty for engaging/disengaging control is small (architecture event) Supports low-power anywhere-anytime of mobile computing Non-timing critical tasks Real-time application that requires more predictable output Concerns i/t during pipeline switch Real-Register File may require extra dedicated ports Bypass bus may have additional loading May, 2002 15

Recently Viewed Presentations

  • Air Force Materiel Command War-Winning Capabilities  On Time,

    Air Force Materiel Command War-Winning Capabilities On Time,

    Be Careful Regarding Scope. An overly broad, all encompassing SOW violates the Competition in Contracting Act (CICA). A SOW that is so broad that it does not reasonably describe the scope of the services needed does not give potential offerors...
  • An Overview of SPICE - Naif

    An Overview of SPICE - Naif

    A SPICE Update(Uses animations: display using PowerPoint's "Slide Show" Mode or "Presenter View") May 2018. Charles Acton. NAIF Manager. The research described in this publication was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract...
  • Mira Loma Seniors - San Juan Unified School District

    Mira Loma Seniors - San Juan Unified School District

    Come to Mira Loma's . Financial Aid Night October 29, 2014. Do NOT pay to fill out your FAFSA. Cal Grants (for students attending college in California): By filling out your FAFSA you become eligible for Cal Grants. We will...
  • Rules of The Game from The Joy Luck Club

    Rules of The Game from The Joy Luck Club

    Is Mrs. Jong a positive force or a negative force in Waverly's development? Do you know anyone who has "invisible strength"? How does it show? Do you have it? If you knew someone who was moving to America, what are...
  • Informal Resources and Constraints Freewrite What formal resources

    Informal Resources and Constraints Freewrite What formal resources

    Informal resources & constraints Skills: Persuasion Hidden hand leadership Personality Ambition Changes in the dominant political rhetoric Richard Neustadt The constitutional presidency is a clerkship Presidential power is the power to persuade Build professional reputation and public prestige in order...
  • Testing new approaches: the Claimant Commitment and local

    Testing new approaches: the Claimant Commitment and local

    Who provides that support? Where does the funding come from? Who manages it all? West Lindsey is the largest and most rural Lincolnshire district within the East Midlands Region at 115,773 hectares (447 square miles). With a population of 89,400,...
  • Inequalities, inequalities…..

    Inequalities, inequalities…..

    Global income inequality today Branko Milanovic Perugia, June 2009 Email: [email protected] Based on the book Worlds Apart, 2005 and updates BM note: this is an update of moscow2.ppt 1. Inequalities today Three concepts of inequality defined Inequality, 1950-2006: The mother...
  • Ib Literature: The Written Assignment

    Ib Literature: The Written Assignment

    The Written Assignment . This is worth 25% of your grade in IB Literature or 25 Marks. It is externally graded. It is a 1200-1500 word essay on one of the works from Part 1 (Literature in Translation) of the...