0. CS/ECE 3330 Computer Architecture
Chapter 1
Power / Parallelism

1. Last Time Performance Analysis It’s all relative
Make sure the units cancel out!
What is a Hz?
Amdahl’s Law
Benchmarking

2. Why Worry about Power Dissipation?
Thermal issues: affect cooling, packaging, reliability, timing
Battery
life
Environment

3. Power Trends
“The Power Wall”

4. Power Dissipation Has Peaked
Must design with strict power envelopes
130W servers, 65W desktop, 10-30W laptops, 1W mobile

5. How Hot Does it Get?

6. Cooling Issues

7. Intel vs. Duracell No Moore’s Law in batteries: 2-3%/year growth
16x
14x
Processor (MIPS)
12x
Improvement (compared to year 0)
Hard Disk (capacity)
10x 8x 6x
Memory (capacity) 4x 2x
Battery (energy stored) 1x
Time (years)
No Moore’s Law in batteries: 2-3%/year growth

8. Environment
Environment Protection Agency (EPA): computers consume 10% of commercial electricity consumption
Includes peripherals, possibly also manufacturing
Data center growth was cited as a contribution to the 2000/2001 California Energy Crisis
Equivalent power (with only 30% efficiency) for AC
CFCs used for refrigeration
Lap burn
Fan noise
Chlorofluorocarbons (CFCs) cited as the cause of ozone depletion

9. Power Matters at Scale…
[J. Koomey
(LBL), 2007]
Eric Schmidt, CEO of Google: "What matters most to the computer designers at Google is not speed, but power - low power, because data centers can consume as much electricity as a city."

10. But Remember Amdahl’s Law

11. Power vs. Energy

12. Power vs. Energy Power consumption in watts
Determines battery life in hours
Sets packaging limits
Energy efficiency in joules
Rate at which power is consumed over time
Energy = power * delay (joules = watts * seconds)
Lower energy number means less power to perform a computation at same frequency

13. Another Fallacy: Low Power at Idle
Morgan Kaufmann Publishers
April 23, 2017
Another Fallacy: Low Power at Idle
X4 power benchmark
At 100% load: 295W
At 50% load: 246W (83%)
At 10% load: 180W (61%)
Google data center
Mostly operates at 10% – 50% load
At 100% load less than 1% of the time
Consider designing processors to make power proportional to load
Chapter 1 — Computer Abstractions and Technology

14. Capacitive Power Dissipation
Capacitance:
Function of wire length, transistor size
Supply Voltage:
Has been dropping with successive fab generations
Power ~ C V2 f
Frequency switched:
Clock frequency + likelihood of change

15. Morgan Kaufmann Publishers
April 23, 2017
Reducing Power
Suppose a new CPU has
75% of capacitive load of old CPU
25% voltage and 25% frequency reduction
The power wall
We can’t reduce voltage further
We can’t remove more heat
How else can we improve performance?
Chapter 1 — Computer Abstractions and Technology

16. Uniprocessor Performance
Morgan Kaufmann Publishers
April 23, 2017
Uniprocessor Performance
Constrained by power, instruction-level parallelism, memory latency
Chapter 1 — Computer Abstractions and Technology

17. Morgan Kaufmann Publishers
April 23, 2017
Multiprocessors
Multicore microprocessors
More than one processor per chip
Multiprocessors and clusters – another course
Requires explicitly parallel programming
Compare with instruction-level parallelism
Hardware executes multiple instructions at once
Hidden from the programmer
Hard to do
Programming for performance
Load balancing
Optimizing communication and synchronization
Traditional multiprocessors and clusters are covered in detail in another course
Chapter 1 — Computer Abstractions and Technology

18. Multicore Architecture Examples
Morgan Kaufmann Publishers
23 April, 2017
Multicore Architecture Examples
2 × quad-core Intel Xeon e5345 (Clovertown)
2 × quad-core AMD Opteron X (Barcelona)
Chapter 7 — Multicores, Multiprocessors, and Clusters

19. Multicore Architecture Examples
Morgan Kaufmann Publishers
23 April, 2017
Multicore Architecture Examples
2 × oct-core Sun UltraSPARC T (Niagara 2)
2 × oct-core IBM Cell QS20
Chapter 7 — Multicores, Multiprocessors, and Clusters

20. Key Points Power has become a limiting factor Power vs energy
P = C * (V^2) * F
One solution: Multicore processors
Different scale than “old” parallel processors
More detail in Chapter 7