Scribd
Upload
2 views

Lecture 3

The document explains Speedup and Amdahl's Law in the context of parallel computing, highlighting how to measure performance improvements. Speedup quantifies the enhancement in execution time with multiple processors, while Amdahl's Law sets theoretical limits on speedup based on the non-parallelizable portions of a program. Key insights include diminishing returns on adding processors and the importance of minimizing sequential code to maximize speedup.

Uploaded by

Naik Amal Shah
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
2 views

Lecture 3

The document explains Speedup and Amdahl's Law in the context of parallel computing, highlighting how to measure performance improvements. Speedup quantifies the enhancement in execution time with multiple processors, while Amdahl's Law sets theoretical limits on speedup based on the non-parallelizable portions of a program. Key insights include diminishing returns on adding processors and the importance of minimizing sequential code to maximize speedup.

Uploaded by

Naik Amal Shah
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 8

Speedup and Amdahl's

Law
• When improving the performance of a system using parallel
computing, it's important to measure how much speed improvement
we gain. Speedup and Amdahl’s Law help us understand the limits of
parallelization.
1. Speedup (S)
Example:

• A program takes 100 seconds on 1 processor.


• The same program takes 25 seconds on 4 processors.
• S(4)=10025=4S(4) = \frac{100}{25} = 4S(4)=25100​=4
• This means the program runs 4 times faster with 4 processors.
2. Amdahl’s Law
• Amdahl’s Law defines the theoretical maximum speedup we can
achieve by parallelizing a program. It considers that some parts of the
program cannot be parallelized.
Example of Amdahl’s Law
Key Insights from Amdahl’s Law
1.Diminishing Returns:
1. Adding more processors does not always mean a proportional speedup.
2. If a large portion of the program is sequential, adding processors won’t help
much.
2.Parallelization Limits:
1. If 90% of a program is parallelizable, even with infinite processors, the
maximum speedup is 10×:
3.Practical Considerations:
1. Beyond a certain number of processors, performance gains are negligible due
to communication overhead and synchronization delays.
Conclusion
• Speedup tells us how much performance improvement we get by
adding processors.
• Amdahl’s Law shows that speedup is limited by the sequential portion
of the code.
• The key to maximizing speedup is minimizing the sequential part and
improving parallel efficiency.

You might also like