This presentation discusses the infamous though old Green Computing field. It also discusses the current and future approaches to achieve more efficient solutions for it.
It was Presented on elective course "Selected Topics in advanced Embedded Systems" at university.
Artificial intelligence in the post-deep learning era
Energy-aware Computing
1. Energy-aware Computing
Software approaches and other technologies
Name ID
Abd ElRahman Abd Elkawy 19-4735
Kareem Rezk 19-9237
Mohamed Elhawary 19-7157
Omar Elshal 19-8014
2. Content Layout:
• Why Energy-aware ?
• Energy and Environment (Green Computing)
• Power reduction Approaches
• Low power through parallelism
• Graphene, how can it contribute to Energy-aware?
• Graphene supercapacitor
• References
3. Why Energy-aware?
• Data centers consumed 61 billion kilowatt-hours (kWh) in 2006 (1.5%
of total U.S. electricity consumption costing $4.5 billion)
• According to Koomey’s report (2011), only 56% increase through
2006-2011 due to virtualization
• 2011- $7.4 billion (25 power plants)
• But still the growth is exponential.
Source: T. Hoefler: Software and Hardware Techniques for Power-Efficient HPC
Networking
4. Why Energy-aware?
• Processors are getting hotter
• Heat needs to be transferred away, or
the chip dies:
• For every 10 degree Celsius increase in
temperature, the lifetime of a chip
reduces by half !
• Expensive solution (liquid cooling)
• Fans, but consume power too
5. Energy and Environment (Green Computing)
• It’s the study of designing, manufacturing, using and disposing of
computers, servers and associated subsystems efficiently and
effectively with minimal on environment.
• According to German Federal Environment office, computers
consume around 17billion kWh each year in standby mode only !
• The CO2 dissipated from ‘sleeping devices’ = 1/7 the CO2 emitted
from a car
7. Algorithmic level
• Fewer instructions/cycles reduces energy
• Trying alternative algorithms with lower complexity:
• E.g. quick sort O(nlogn) , bubble sort O(n^2)
• Heuristic approach, go for a good solution, but not the best
• Biggest gains at this level
8. Compiler level
• Strength reduction
• E.g. replace multiplications with Add’s and Shift’s
• E.g. replace floating point with fixed point
• Source-to-Source transformation
• Loop transformation to improve locality
• Reorder instructions to reduce bit-transition
• Reduce register pressure (number of accesses to register file)
• Perform special optimizations per scenario of each execution mode
9. Architecture, Organization level
• Going parallel
• Add local memories
For Organization level (micro Architecture)
• Reducing Vdd by using lower freq.
• Pipelining(cheap way of parallelism)
• Reducing register traffic
• Avoid unnecessary reads and writes
10. Circuit level
• Clock gating
• add more logic to circuit to prune clock tree
• Power gating
• shut off current to blocks not in use in circuit
• Use special SRAM cells
• Normal SRAM can’t scale below Vdd =0.7-0.8 Volt
• Multiple Vdd modes
11. Low power through parallelism
• Sequential Processor
• Switching capacitance C
• Frequency f
• Voltage V
• P1 = αfCV2
• Parallel Processor (two times the number of units)
• Switching capacitance 2C
• Frequency f/2
• Voltage V’ < V
• P2 = αf/2*2CV’2 = αfCV’2 < P1
12. Graphene & Energy-aware
• It’s a single layer of graphite (pure crystalline carbon)
• First isolated in lab in 2004 by Andre Geim and Konstantin Novoselov at the
University of Manchester (won Nobel Prize in Physics in 2010)
• It is the thinnest material imaginable (~0.345 nm thick)
• It is electrically conductive – best known so far
• 1,000,000x more conductive than copper (current density at room temp.)
• Replacement for Solar cells, touchscreens, new computers, batteries, etc.
13. Graphene supercapacitor
• The graphene supercapacitor is capable of charging up to 1,000 times faster than
a normal battery
• Fully charge your phone in 30 seconds and last for days
• Contains no toxic chemical, carbon based (Green)
• Ten grams of graphene is the same weight as two nickels
• These ten grams could cover the electricity of Cairo stadium
• 15 Kgs of graphene would cover all of the computer displays in the world
• 15 kgs of graphene is equivalent in weight to a standard cinder block
clock gating saves power by adding more logic to a circuit to prune the clock tree. Pruning the clock disables portions of the circuitry so that the flip-flops in them do not have to switch states. Switching states consumes power. When not being switched, the switching power consumption goes to zero