Академический Документы
Профессиональный Документы
Культура Документы
In some situations, you might be told up front that you will have 3W
available to run your design.
0.9
Utilization 1
0.8 Leakage limited scaling
0.7
Device count S2
0.6
0.5
Device frequency S
0.4 Device power (cap) 1/S
0.3 Device power (Vdd) ~1
0.2
Utilization 1/S2
0.1
0.0
90nm 65nm 45nm 32nm
The Utilization Wall
2
Utilization @ 300mm & 80w
Experimental results
0.20
Replicated small data path 17.6%
0.18
More Dark Silicon than active
0.16
0.14
0.12
Observations in the wild 0.10
Flat frequency curve
0.08
Turbo Mode 6.5%
0.06
Increasing cache/processor ratio 3.3%
0.04
0.02
0.00
90nm 45nm 32nm
TSMC TSMC ITRS
Utilization Wall: Dark Implications for Multicore
.
Spectrum of tradeoffs
between # cores and
frequency.
2x4 cores @ 3 GHz
e.g.; take (8 cores dark)
65 nm32 nm; (Industrys Choice)
i.e. (s =2)
.
4 cores @ 3 GHz
.
65 nm 32 nm
Key Insights
The research leverages two key insights:
Energy efficient
Up to 16x for targeted hot code
Cold code
Figure shows the projected energy
savings in GreenDroid and the origin
of these savings.
The savings come from two sources
First, c-cores dont require
instruction fetch, instruction decode,
a conventional register file, or any of
the associated structures. Removing
these reduces energy consumption
by 56 percent.
The second source of savings (35 %
of energy) comes from the
specialization of the c-cores data
path.
The result is that average per-
instruction energy drops from 91 pJ
per instruction to just 8 pJ per
instruction.
Marching Toward Completion
The toolchain automatically generates placed-and-routed c-core
tiles, given the source code and information about execution
properties.
The cycle- and energy-accurate simulation tools have confirmed the
energy savings provided by c-cores.
The team is currently working on more detailed full-system Android
emulation to improve our workload modeling so that they can finalize
the selection of c-cores that will populate GreenDroids dark silicon.
In parallel with this effort, they are working ontiming closure and
physical design.
The Research Team
The University of California,San Diego,USA. Massachusetts Institute of
Technology (MIT),USA
Assistant professors PhD students Postdoctoral Researcher
Michael Bedford Taylor Nathan Goulding-Hotta Jonathan Babb
Ganesh Venkatesh
Steven Swanson
Saturnino Garcia
Postdoctoral scholar Manish Arora
Jack Sampson Siddhartha Nath
Graduate Students
Vikram Bhatt
Joe Auricchio
Po-Chao Huang
Conclusions
Over the next five to 10 years, the breakdown of conventional silicon
scaling and the resulting utilization wall will exponentially increase
the amount of dark silicon in both desktop and mobile processors.
The GreenDroid prototype demonstrates that c-cores offer a new
technique to convert dark silicon into energy savings and increased
parallel execution under strict power budgets.
The estimate that the prototype will reduce processor energy
consumption by 91 percent for the code that c-cores target, and
result in an overall savings of 7.4 X.
22
Reference
Based on April 2011 IEEE Micro paper.