Click here to load reader

Accordion - Electrical and Computer ukarpuzc/Karpuzcu_files/accordionTalk.pdf · PDF fileAccordion Accordion Basics •How to close the gap between NTC and STC execution times?!5

  • View
    296

  • Download
    6

Embed Size (px)

Text of Accordion - Electrical and Computer ukarpuzc/Karpuzcu_files/accordionTalk.pdf · PDF...

  • Accordion Toward Soft Near-threshold Voltage Computing

    Ulya R. Karpuzcu, Ismail Akturk Nam Sung Kim

  • Accordion

    Near-threshold Voltage Computing (NTC)

    !2

  • Accordion

    Near-threshold Voltage Computing (NTC)Supply voltage Vdd remains slightly above threshold voltage Vth

    !2

  • Accordion

    Near-threshold Voltage Computing (NTC)Supply voltage Vdd remains slightly above threshold voltage Vth

    !2

    VddMIP

    S/W

    att

    VddSTVVddNTVVth

  • Accordion

    Near-threshold Voltage Computing (NTC)Supply voltage Vdd remains slightly above threshold voltage VthEnergy efficiency increases as Vdd reaches Vth

    !2

    VddMIP

    S/W

    att

    VddSTVVddNTVVth

  • Accordion

    Near-threshold Voltage Computing (NTC)Supply voltage Vdd remains slightly above threshold voltage VthEnergy efficiency increases as Vdd reaches Vth

    2-4x more energy efficient than super-threshold (STV) operation

    !2

    VddMIP

    S/W

    att

    VddSTVVddNTVVth

    2x-4x

  • Accordion

    Near-threshold Voltage Computing (NTC)Supply voltage Vdd remains slightly above threshold voltage VthEnergy efficiency increases as Vdd reaches Vth

    2-4x more energy efficient than super-threshold (STV) operation

    !2

    VddMIP

    S/W

    att

    VddSTVVddNTVVth

    2x-4x

    How close to Vdd can Vth get?

  • Accordion !3

    NTC Basics: How close to Vdd can Vth get?

  • Accordion !3

    NTC Basics: How close to Vdd can Vth get?

    Vdd

    log(

    f)

    VddSTVVddNTVVth

  • Accordion !3

    NTC Basics: How close to Vdd can Vth get?

    5x-10x

    Vdd

    log(

    f)

    VddSTVVddNTVVth

  • Accordion !3

    NTC Basics: How close to Vdd can Vth get?

    5x-10x

    Vdd

    log(

    f)

    VddSTVVddNTVVth

    Execution time is proportional to work per parallel task x f

  • Accordion !3

    NTC Basics: How close to Vdd can Vth get?

    5x-10x

    Vdd

    log(

    f)

    VddSTVVddNTVVth

    Execution time is proportional to work per parallel task x fNo degradation if 5-10x more cores engaged in computation

  • Accordion !3

    NTC Basics: How close to Vdd can Vth get?Po

    wer

    Vdd

    5x-10x

    Vdd

    log(

    f)

    VddSTVVddNTVVth

    Execution time is proportional to work per parallel task x fNo degradation if 5-10x more cores engaged in computation

  • Accordion !3

    NTC Basics: How close to Vdd can Vth get?Po

    wer

    Vdd

    10x-40x

    5x-10x

    Vdd

    log(

    f)

    VddSTVVddNTVVth

    Execution time is proportional to work per parallel task x fNo degradation if 5-10x more cores engaged in computation10-40x power savings per core can accommodate 5-10x more cores

  • Accordion !3

    NTC Basics: How close to Vdd can Vth get?Po

    wer

    Vdd

    10x-40x

    5x-10x

    Vdd

    log(

    f)

    VddSTVVddNTVVth

    Execution time is proportional to work per parallel task x fNo degradation if 5-10x more cores engaged in computation10-40x power savings per core can accommodate 5-10x more cores

    Limited by the degree of parallelism in application

  • Accordion !4

    NTC Basics: How close to Vdd can Vth get?

    0.45 0.50 0.55 0.60

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    Vdd (V)

    Tim

    ing

    Erro

    r Rat

    e

    0.45 0.50 0.55

    0.0

    0.4

    Vdd (V)

    Timi

    ng E

    rror

    Rat

    e

    0.2

    0.6

    0.81.

    0

    0.60

  • Accordion !4

    NTC Basics: How close to Vdd can Vth get?

    0.45 0.50 0.55 0.60

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    Vdd (V)

    Tim

    ing

    Erro

    r Rat

    e

    0.45 0.50 0.55

    0.0

    0.4

    Vdd (V)

    Timi

    ng E

    rror

    Rat

    e

    0.2

    0.6

    0.81.

    0

    0.60

  • Accordion !4

    NTC Basics: How close to Vdd can Vth get?

    0.45 0.50 0.55 0.60

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    Vdd (V)

    Tim

    ing

    Erro

    r Rat

    e

    0.45 0.50 0.55

    0.0

    0.4

    Vdd (V)

    Timi

    ng E

    rror

    Rat

    e

    0.2

    0.6

    0.81.

    0

    0.60

    Limited by the degree of vulnerability to variation

  • Accordion

    Accordion Basics

    !5

  • Accordion

    Accordion BasicsHow to close the gap between NTC and STC execution times?

    !5

  • Accordion

    Accordion BasicsHow to close the gap between NTC and STC execution times?

    !5

    Execution Time / Problem Sizef Core Count

  • Accordion

    Accordion BasicsHow to close the gap between NTC and STC execution times?

    !5

    Problem SizeNTV

    fNTV Core CountNTV! Problem SizeSTV

    fSTV Core CountSTV

    Execution Time / Problem Sizef Core Count

  • Accordion

    Accordion BasicsHow to close the gap between NTC and STC execution times?

    !5

    Problem SizeNTV

    fNTV Core CountNTV! Problem SizeSTV

    fSTV Core CountSTV

    Execution Time / Problem Sizef Core Count

    fNTV < fSTV

  • Accordion

    Accordion BasicsHow to close the gap between NTC and STC execution times?

    !5

    Problem SizeNTV

    fNTV Core CountNTV! Problem SizeSTV

    fSTV Core CountSTV

    Execution Time / Problem Sizef Core Count

    fNTV < fSTV Core CountNTV > Core CountSTV

  • Accordion

    Accordion BasicsHow to close the gap between NTC and STC execution times?

    !5

    Designate the problem size as the main knob to adjust

    Problem SizeNTV

    fNTV Core CountNTV! Problem SizeSTV

    fSTV Core CountSTV

    Execution Time / Problem Sizef Core Count

    fNTV < fSTV Core CountNTV > Core CountSTV

  • Accordion

    Accordion BasicsHow to close the gap between NTC and STC execution times?

    !5

    Designate the problem size as the main knob to adjustthe degree of parallelism

    Problem SizeNTV

    fNTV Core CountNTV! Problem SizeSTV

    fSTV Core CountSTV

    Execution Time / Problem Sizef Core Count

    fNTV < fSTV Core CountNTV > Core CountSTV

  • Accordion

    Accordion BasicsHow to close the gap between NTC and STC execution times?

    !5

    Designate the problem size as the main knob to adjustthe degree of parallelismthe degree of vulnerability to variation

    Problem SizeNTV

    fNTV Core CountNTV! Problem SizeSTV

    fSTV Core CountSTV

    Execution Time / Problem Sizef Core Count

    fNTV < fSTV Core CountNTV > Core CountSTV

  • Accordion

    Problem Size vs. Quality of Computing

    !6

  • Accordion

    Problem Size vs. Quality of Computing

    !6

    Problem Size

    1.1

    1.0

    0.5 1.0 1.5 2.0

    0.9

    0.8

    Qual

    ity

    Default

  • Accordion

    Problem Size vs. Quality of Computing

    !7

    Problem Size

    1.1

    1.0

    0.5 1.0 1.5 2.0

    0.9

    0.8

    Qual

    ity

    Drop1/4Default

  • Accordion

    Problem Size vs. Quality of Computing

    !8

    Problem Size

    1.1

    1.0

    0.5 1.0 1.5 2.0

    0.9

    0.8

    Qual

    ity

    Drop1/2Drop1/4Default

  • Accordion

    Problem Size vs. Quality of Computing

    !8

    Problem Size

    1.1

    1.0

    0.5 1.0 1.5 2.0

    0.9

    0.8

    Qual

    ity

    Drop1/2Drop1/4Default

    Execution Time

    / Problem Sizef Core Count

  • Accordion

    Problem Size vs. Quality of Computing

    !8

    Problem Size

    1.1

    1.0

    0.5 1.0 1.5 2.0

    0.9

    0.8

    Qual

    ity

    Drop1/2Drop1/4Default

    Execution Time

    / Problem Sizef Core Count

  • Accordion

    Problem Size vs. Quality of Computing

    !8

    Problem Size

    1.1

    1.0

    0.5 1.0 1.5 2.0

    0.9

    0.8

    Qual

    ity

    Drop1/2Drop1/4Default

    Execution Time

    / Problem Sizef Core Count

    To make up for f degradation:

  • Accordion

    Problem Size vs. Quality of Computing

    !8

    Problem Size

    1.1

    1.0

    0.5 1.0 1.5 2.0

    0.9

    0.8

    Qual

    ity

    Drop1/2Drop1/4Default

    Execution Time

    / Problem Sizef Core Count

    To make up for f degradation:Compress problem size

  • Accordion

    Problem Size vs. Quality of Computing

    !8

    Problem Size

    1.1

    1.0

    0.5 1.0 1.5 2.0

    0.9

    0.8

    Qual

    ity

    Drop1/2Drop1/4Default

    Execution Time

    / Problem Sizef Core Count

    To make up for f degradation:Compress problem sizeExpand problem size

  • Accordion

    Problem Size vs. Quality of Computing

    !8

    Problem Size

    1.1

    1.0

    0.5 1.0 1.5 2.0

    0.9

    0.8

    Qual

    ity

    Drop1/2Drop1/4Default

    Execution Time

    / Problem Sizef Core Count

    To make up for f degradation:Compress problem sizeExpand problem size

    Core Count should expand

  • Accordion

    Variation Induced Errors

    !9

    0.45 0.50 0.55 0.60

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    Vdd (V)

    Tim

    ing

    Erro

    r Rat

    e

    0.45 0.50 0.55

    0.0

    0.4

    Vdd (V)

    Timi

    ng E

    rror

    Rat

    e

    0.2

    0.6

    0.81.

    0

    0.60

  • Accordion

    Variation Induced Errors

    !9

    0.45 0.50 0.55 0.60

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    Vdd

Search related