Transcript Computer Architecture:

High Performance Computer
Architecture:
Introduction:
What is Computer Architecture:
Where does Architecture stand?
Levels of Transformation:
Power of Abstraction:
An example: Multi Core Systems
No of transistors on ICs double every year:
Crossing the Abstraction Layers:
Why do we need Computer Architecture:
Simple Computer Architecture Quiz:
Technology Trends
Good Old Moore’s Law (Technology vs
Architects):
Speed Doubling Quiz:
Quiz Solution:
Impact of Moore on Speed : Memory Wall:
Processor Performance Metrics:
Speed vs Power vs Weight vs Cost
Power Consumption in a Processor:
Active Power:
• Capacitance(C) : direct function of chip area. More Area = more C.
• Technology improvements leads to lesser area, incorporate multiple
processors to make it more powerful with same power dissipation.
• Smaller Transistors are faster, increase in frequency, leading to
greater power dissipation.
• Smaller Transistors lower the supply voltage maintaining the same
speed.
• Too small a voltage also doesn’t lead to optimum frequencies.
• Can’t increase frequency too much, leads to higher voltage levels.
• Doubling the Supply Voltage will Quadruple the Power Supply.
Active Power Analysis:
Static Power : Analogy with Water Valve
• It prevents lowering the voltage too much.
• Consider circuits like tap, although the valve is closed, The pressure(V)
is controlling the valve.
• In a transistor the valve is controlled by water pressure from another
valve (V)
• If the valve is closed and the pressure applied is not enough(V), water
will leak (I)
• Therefore leakage current increases when Voltage(Water Pressure is
reduced)
The Optimum Power consumed should be
somewhere in between the graph:
Wafer Cost Reduction:
• The number of good chips being manufactured/fabricated is a key
factor for an organization’s revenue.
• Smaller sized working chips give a higher yield % as they occupy
lesser area.
• Wafer might have defects due to impure Silicon or impurity caused
by the fabrication steps.
Fabrication Yield Example:
More Bang for your Buck:
Metrics and Evaluation:
Parameters for Metrics and Evaluation:
• What does better mean in Computer Architecture? Is the speed
(GHz) or the Memory size(GB)?
• Latency and Throughput are two key performance parameters.
• Latency: time taken from start to end for a process
• Throughput: Number of computations per second (#/second)
Performance:
Comparing CPU/Processor Performance:
Speed Up:
Measuring Performance:
Measuring Performance(Inaccurate Way):
Benchmarks (Accurate Way of Measuring
Performance):
Summarizing Performance(Average Execution
Time Preferred over Average Speedup):
Summarizing Performance II (Averaging
Speedups not preferred):
• Speed up of average Execution time not the same as average of
speed ups of the individual applications.
• Geometric Mean can be used to average speed ups
Iron Law of Performance in CPU:
Iron’s Law for Performance – II :
Diminishing Returns on 1-Dimensional
Improvement: