penn ese534 spring2014 -- dehon ese534: computer organization day 15: march 24, 2014 empirical...
TRANSCRIPT
![Page 1: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/1.jpg)
Penn ESE534 Spring2014 -- DeHon
ESE534:Computer Organization
Day 15: March 24, 2014
Empirical Comparisons
![Page 2: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/2.jpg)
Penn ESE534 Spring2014 -- DeHon
Previously
• Instruction Space Modeling
![Page 3: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/3.jpg)
Previously
• Programmable compute blocks– LUTs, ALUs, PLAs
Penn ESE534 Spring2014 -- DeHon
![Page 4: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/4.jpg)
Today
• What if we just built a custom circuit?– What cost are we paying for programmability?
• Can we afford to build custom circuits?– Can we afford not to?
• Empirically compare artifacts
• Different kind of lecture – messier, real-world artifacts
• Coming at this about 3 different ways– Bottom up from sizes; with full benchmarks; particular
examples
Penn ESE534 Spring2014 -- DeHon
![Page 5: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/5.jpg)
Penn ESE534 Spring2014 -- DeHon
Today
• Empirical Data– Custom
• Gate Array• Std. Cell (ASIC)• Full
– FPGAs– Processors– NRE– Tasks
![Page 6: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/6.jpg)
Preclass 1
• How big?– 2-LUT?– 2-LUT w/ Flip-flop?– 2-LUT w/ 4 input sources?– 2-LUT w/ 200 input sources?
Penn ESE534 Spring2014 -- DeHon
![Page 7: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/7.jpg)
Penn ESE534 Spring2014 -- DeHon
Empirical Comparisons
![Page 8: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/8.jpg)
Penn ESE534 Spring2014 -- DeHon
Empirical
• Ground modeling in some concretes
• Start sorting out– custom vs. configurable– spatial configurable vs. temporal
• Start by reviewing alternatives…
![Page 9: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/9.jpg)
Penn ESE534 Spring2014 -- DeHon
Full Custom
• Get to define all layers
• Use any geometry you like
• Only rules are process design rules
• ESE570
![Page 10: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/10.jpg)
Penn ESE534 Spring2014 -- DeHon
Standard Cell Area
inv nand3 AOI4inv nor3 Inv
All cellsuniformheight
Width ofchanneldeterminedby routing
Cell area
![Page 11: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/11.jpg)
Penn ESE534 Spring2014 -- DeHon
Standard Cell Area
inv nand3 AOI4inv nor3 Inv
All cellsuniformheight
Width ofchanneldeterminedby routing
Cell area
What freedom have we removed? Impact?
![Page 12: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/12.jpg)
Penn ESE534 Spring2014 -- DeHon
MPGA• Metal Programmable Gate Array
– Resurrected as “Structured ASICs”
• Gates pre-placed (poly, diffusion)• Only get to define metal connections
– Today’s structured ASICs – maybe just vias– Cheap (low NRE)
• only have to pay for metal mask(s)
[Wu&Tsai/ISPD2004p103]
![Page 13: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/13.jpg)
Structured ASIC: eASIC
• 2011
• 45nm
• 1.4M LUTs
• 500MHz?
Penn ESE534 Spring2014 -- DeHon
http://www.easic.com/wp-content/uploads/2011/02/eASIC-Nextreme-2T-Product-Brief.pdf
![Page 14: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/14.jpg)
Structured ASIC
• Maybe think about it as an FPGA with vias instead of configurable switches?
• Ratio of SRAM to via design?
Penn ESE534 Spring2014 -- DeHon
![Page 15: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/15.jpg)
What do we expect?
• Comparing density/delay/energy– Full custom– Standard Cell (ASIC)– MPGA / Structured ASIC– FPGA– Processor
Penn ESE534 Spring2014 -- DeHon
![Page 16: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/16.jpg)
Why it isn’t trivial?
• Different logic forms
• Interconnect
• Balance of resources
• Mix of requirements in tasks
Penn ESE534 Spring2014 -- DeHon
![Page 17: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/17.jpg)
Penn ESE534 Spring2014 -- DeHon
MPGA vs. Custom?
• AMI CICC’83– MPGA 1.0– Std-Cell 0.7– Custom 0.5
• AMI CICC’04– Custom 0.6 (DSP)– Custom 0.8 (DPath)
• Toshiba DSP– Custom 0.3
• Mosaid RAM– Custom 0.2
• GE CICC’86– MPGA 1.0– Std-Cell 0.4--0.7
• FF/counter 0.7• FullAdder 0.4• RAM 0.2
![Page 18: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/18.jpg)
Penn ESE534 Spring2014 -- DeHon
Metal Programmable Gate Arrays
![Page 19: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/19.jpg)
Penn ESE534 Spring2014 -- DeHon
MPGAs
• Modern -- “Sea of Gates”
• yield 35--70%
• Maybe 1.25F2/gate ? – (quite a bit of variance)
![Page 20: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/20.jpg)
Penn ESE534 Spring2014 -- DeHon
Conventional FPGA Tile
K-LUT (typical k=4) w/ optional output Flip-Flop
![Page 21: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/21.jpg)
Penn ESE534 Spring2014 -- DeHon
Toronto FPGA Model
![Page 22: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/22.jpg)
Penn ESE534 Spring2014 -- DeHon
FPGA Table
![Page 23: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/23.jpg)
Penn ESE534 Spring2014 -- DeHon
(semi) Modern FPGAs
• APEX 20K1500E 52K LEs 0.18m 24mm 22mm
300KF2/LE
• XC2V1000 10.44mm x 9.90mm
[source: Chipworks]
0.15m 11,520 4-LUTs
1. 5M2/4-LUT (~375KF2/4-LUT)
[Both also have RAM in cited area]
![Page 24: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/24.jpg)
Penn ESE534 Spring2014 -- DeHon
How many gates?(Prelcass 3)
![Page 25: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/25.jpg)
Penn ESE534 Spring2014 -- DeHon
“gates” in 2-LUT
![Page 26: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/26.jpg)
Penn ESE534 Spring2014 -- DeHon
Now how many?
![Page 27: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/27.jpg)
Penn ESE534 Spring2014 -- DeHon
Which gives: Higher fraction of gates used ?
More gates/unit area?
More usable gates?
![Page 28: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/28.jpg)
Penn ESE534 Spring2014 -- DeHon
Gates Required?
Depth=3, Depth=2048?
![Page 29: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/29.jpg)
Penn ESE534 Spring2014 -- DeHon
Gate metric for FPGAs?
• Day11: several components for computations– compute element
– interconnect:• space• time
– instructions
• Not all applications need in same balance
• Assigning a single “capacity” number to device is an oversimplification
![Page 30: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/30.jpg)
Penn ESE534 Spring2014 -- DeHon
MPGA vs. FPGA
• MPGA (SOG GA)– 1.25KF2 /gate– 35-70% usable
(50%)– 1.5-4KF2/gate net
• Ratio: 2--10 (5)
• Xilinx XC4K– 300KF2 /CLB– 17--48 gates (26?)– 6-18KF2/gate net
Adding ~2x Custom/MPGA, Custom/FPGA ~10x
![Page 31: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/31.jpg)
Penn ESE534 Spring2014 -- DeHon
http://www.easic.com/high-speed-transceivers-low-cost-power-fpga-nre-asic-45nm-easic-nextreme-2/easic-nextreme-2-look-up-table-lut-architecture/
![Page 32: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/32.jpg)
FPGA vs. Structure ASIC
• Virtex 6 • 40nm• 470K 6-LUTs• Largest device
• eASIC• 45nm• 580K eCells• Probably smaller die
Penn ESE534 Spring2014 -- DeHon
http://www.easic.com/high-speed-transceivers-low-cost-power-fpga-nre-asic-45nm-easic-nextreme-2/easic-nextreme-2-look-up-table-lut-architecture/
![Page 33: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/33.jpg)
Penn ESE534 Spring2014 -- DeHon
FPGA vs. Std Cell
• 90nm
• FPGA: Stratix II
• STMicro CMOS090– Standard Cell
• Full custom layout• …but by tool
[Kuon/Rose TRCADv26n2p203--215 2007]
![Page 34: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/34.jpg)
Penn ESE534 Spring2014 -- DeHon
MPGA vs. FPGA (Delay)
• MPGA (SOG GA) Fgd~1ns
• Ratio: 1--7 (2.5)– Altera claiming 2×
• For their Structured ASIC [2007]
– LSI claiming 3ו 2005
• Xilinx XC4KF1-7 gates in 7ns
2-3 gates typical
![Page 35: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/35.jpg)
Penn ESE534 Spring2014 -- DeHon
• 90nm
• FPGA: Stratix II
• STMicro CMOS090
[Kuon/Rose TRCADv26n2p203--215 2007]
FPGA vs. Std. CellDelay
![Page 36: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/36.jpg)
Penn ESE534 Spring2014 -- DeHon
• 90nm
• FPGA: Stratix II
• STMicro CMOS090
• eASIC (MPGA) claim– 20% of FPGA power– (best case)
[Kuon/Rose TRCADv26n2p203--215 2007]
FPGA vs. Std CellEnergy
![Page 37: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/37.jpg)
Penn ESE534 Spring2014 -- DeHon
Processors vs. FPGAs
![Page 38: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/38.jpg)
Penn ESE534 Spring2014 -- DeHon
Processors and FPGAs
![Page 39: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/39.jpg)
Penn ESE534 Spring2014 -- DeHon
Component Example
XC4085XL-09 3,136 CLBs 4.6ns
682 Bit Ops/ns
Alpha 1996 264b ALUs 2.3ns
55.7 Bit Ops/ns
• Single die in 0.35m
[1 “bit op” = 2 gate evaluations]
![Page 40: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/40.jpg)
Penn ESE534 Spring2014 -- DeHon
Processors and FPGAs
![Page 41: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/41.jpg)
Penn ESE534 Spring2014 -- DeHon
Raw Density Summary
• Area– MPGA 2-3x Custom– FPGA 5x MPGA
• FPGA:std-cell custom ~ 15-30x
• Area-Time– Gate Array 6-10x Custom– FPGA 15-20x Gate Array
• FPGA:std-cell custom ~ 100x
– Processor 10x FPGA
![Page 42: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/42.jpg)
Penn ESE534 Spring2014 -- DeHon
Raw Density Caveats
• Processor/FPGA may solve more specialized problem
• Problems have different resource balance requirements– …can lead to low yield of raw density
![Page 43: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/43.jpg)
Penn ESE534 Spring2014 -- DeHon
Challenge: NRE
![Page 44: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/44.jpg)
Penn ESE534 Spring2014 -- DeHon
NRE Costs
• 28-nm SoC development costs doubled over previous node – EE Times 201328nm+78%, 20nm+48%, 14nm+31%, 10nm+35%
![Page 45: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/45.jpg)
Penn ESE534 Spring2014 -- DeHon
Economics
• Economics force fewer, more customizable chips– Mask costs approaching millions of dollars– Custom IC design NRE tens of millions of dollars
• Need market of hundreds of millions of dollars to recoup investment
• With fixed or slowly growing total IC industry revenues Number of unique chips must decrease
Forcing fewer,more customizablechips
![Page 46: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/46.jpg)
Penn ESE534 Spring2014 -- DeHon
Task Comparisons
![Page 47: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/47.jpg)
Penn ESE534 Spring2014 -- DeHon
Broadening Picture
• Compare larger computations
• For comparison– throughput density metric: results/area-time
• normalize out area-time point selection• high throughput density
most in fixed area
least area to satisfy fixed throughput target
![Page 48: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/48.jpg)
Penn ESE534 Spring2014 -- DeHon
Multiply
![Page 49: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/49.jpg)
Preclass 4
• Efficiency of 8×8 multiply on 16×16 multiplier?
Penn ESE534 Spring2014 -- DeHon
![Page 50: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/50.jpg)
Penn ESE534 Spring2014 -- DeHon
Multiply
![Page 51: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/51.jpg)
Penn ESE534 Spring2014 -- DeHon
Example: FIR Filtering
Application metric: TAPs = filter taps multiply accumulate
Yi=w1xi+w2xi+1+...
![Page 52: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/52.jpg)
Mixed Designs
• Modern FPGAs include hardwired multipliers (Virtex 25x18)
Penn ESE534 Spring2014 -- DeHonhttp://www.xilinx.com/products/silicon-devices/fpga/virtex-6/index.htm
![Page 53: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/53.jpg)
Penn ESE534 Spring2014 -- DeHon
FPGA vs. Std Cell(revisit)
• 90nm
• FPGA: Stratix II
• STMicro CMOS090
[Kuon/Rose TRCADv26n2p203--215 2007]
![Page 54: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/54.jpg)
Penn ESE534 Spring2014 -- DeHon
Energy
[Abnous et al, The Application of Programmable DSPs in Mobile Communications, Wiley, 2002, pp. 327-360 ]
Pleiades includes hardwire multiply accumulator
![Page 55: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/55.jpg)
Penn ESE534 Spring2014 -- DeHon
• 90nm
• FPGA: Stratix II
• STMicro CMOS090
[Kuon/Rose TRCADv26n2p203--215 2007]
FPGA vs. Std CellEnergy (revisit)
![Page 56: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/56.jpg)
Penn ESE534 Spring2014 -- DeHon
Degrade from Peak
![Page 57: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/57.jpg)
How do various architecture degrade from peak?
• FPGA?
• Processor?
• Custom?
Penn ESE534 Spring2014 -- DeHon
![Page 58: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/58.jpg)
Penn ESE534 Spring2014 -- DeHon
Degrade from Peak: FPGAs
• Long path length not run at cycle
• Limited throughput requirement– bottlenecks elsewhere limit throughput req.
• Insufficient interconnect
• Insufficient retiming resources (bandwidth)
![Page 59: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/59.jpg)
Penn ESE534 Spring2014 -- DeHon
Degrade from Peak: Processors
• Ops w/ no gate evaluations (interconnect)
• Ops use limited word width
• Stalls waiting for retimed data
![Page 60: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/60.jpg)
Penn ESE534 Spring2014 -- DeHon
Degrade from Peak: Custom/MPGA
• Solve more general problem than required– (more gates than really need)
• Long path length
• Limited throughput requirement
• Not needed or applicable to a problem
![Page 61: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/61.jpg)
Penn ESE534 Spring2014 -- DeHon
Degrade Notes
• We’ll cover these issues in more detail as we get into them later in the course
![Page 62: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/62.jpg)
Penn ESE534 Spring2014 -- DeHon
Big Ideas[MSB Ideas]
• Raw densities: custom:ga:fpga:processor– 1:5:100:1000– close gap with specialization
![Page 63: Penn ESE534 Spring2014 -- DeHon ESE534: Computer Organization Day 15: March 24, 2014 Empirical Comparisons](https://reader035.vdocument.in/reader035/viewer/2022062515/56649f505503460f94c724c8/html5/thumbnails/63.jpg)
Penn ESE534 Spring2014 -- DeHon
Admin
• Grading HW5.2 (not touched, yet)
• HW6 due Wednesday
• HW7 out
• Wednesday Reading on web– Classic Paper