36
Accelerator Architectures: High-Level Modeling of Specialization 3% ,**&- $**' * )#%),%)# ) ++'% %)- ,3, )%3,-%.6

Brooks MODSIM August 2015 - PNNL · 2015. 8. 13. · pp l l l l l l l '$+& ), (/ / ! b5 mbg5 ll cb5 ll cc5 ! cd5 ! ce5 l cf5 '( i5 ! j5 l k5 '( c5 ! d5 ! e5 l f5 '( cg5 ll ch5 ! ci5

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

  • Accelerator Architectures:High-Level Modeling of Specialization

  • [Brodersen and Meng, 2002]

    16 Encryption

    17 Hearing Aid

    18 FIR for disk read

    19 MPEG Encoder

    20 802.11 Baseband

  • • – 

    • – 

    • – 

  • 0 200 400 600 800 1000 1200Execution Time (uS)

    0

    20

    40

    60

    80

    100

    120

    140

    Pow

    er(m

    W)

    HLS

  • 0 200 400 600 800 1000 1200Execution Time (uS)

    0

    20

    40

    60

    80

    100

    120

    140

    Pow

    er(m

    W)

    ALADDINHLS

  •           

  • • • 

    – – – 

    • – 

    • – – – 

    • – 

  • • 

    • – – 

    • – – – – 

  • • 

    • 

    • – – 

  • CPU

    DMA Engine

    Scratchpad

    TLB

    DRAM

    LLC

    CacheCache

    Acc Datapath

  • – 

  • 0 200 400 600 800 1000 1200Execution Time (uS)

    0

    20

    40

    60

    80

    100

    120

    140

    Pow

    er(m

    W)

    ALADDINHLS

    – 

    100

    W)

    20

    40

    60Po

  • Harvard Robobee Brain SoC

    “Datacenter Tax” prime candidates for

    specialization

  • • – – 

    • 

    –