2 minute read

What is Ahmdahl’s Law?

Amdahl’s Law, formulated by Gene Amdahl in 1967, guides performance optimization in computer systems. It quantifies the potential speedup achievable through parallelization and optimization. In this blog post, we explore Amdahl’s Law, its implications, and how it can unlock computing system performance.

Understanding Amdahl’s Law

Amdahl’s Law determines the maximum speedup possible based on the proportion of a program that can be parallelized. It is expressed as:

Speedup = 1 / [(1 - P) + (P / N)]


  • Speedup: Performance improvement from optimizing a portion of the program.
  • P: Proportion of the program that can be parallelized.
  • N: Number of processors or threads available for parallel execution.

Key Implications

  1. Limitations of Parallelization: Maximum speedup is limited by the non-parallelizable portion of the program.
  2. Focusing on Critical Sections: Optimizing critical sections that consume significant execution time maximizes potential speedup.
  3. Balancing Parallelization and Optimization: Combine parallelization and optimization efforts for optimal results.
  4. Importance of Scalability: Consider scalability to avoid diminishing returns from excessive parallelization.

Applying Amdahl’s Law for Performance Optimization

  1. Profiling and Analysis: Profile and analyze the program to identify critical sections for targeted optimization.
  2. Parallelization Strategies: Use threading or task-based parallelism to exploit available resources efficiently.
  3. Optimization Techniques: Employ algorithmic improvements, caching strategies, data structure optimizations, and compiler optimizations for non-parallelizable sections.
  4. Measurement and Iteration: Continuously measure and evaluate performance improvements, refining the approach iteratively.

Example: Image Processing Application

Let’s consider an image processing application that applies a series of filters to enhance and transform images. The application consists of three main tasks: image loading (T1), filtering (T2), and image saving (T3).

Upon profiling the application, we find that 70% of the execution time is spent on the filtering task (T2), which can be parallelized, while the remaining 30% is evenly distributed between image loading (T1) and image saving (T3), which are sequential tasks.

Suppose we have access to a multi-core system with eight processors for parallel execution. By applying Amdahl’s Law, we can estimate the potential speedup achievable by parallelizing the filtering task (T2).

Using Amdahl’s Law, we get:

Speedup = 1 / [(1 - 0.7) + (0.7 / 8)] = 1 / (0.3 + 0.0875) = 1 / 0.3875 = 2.58.

The result indicates that we can potentially achieve a speedup of 2.58 times by parallelizing the filtering task (T2) on the available eight processors. This means the image processing application can run approximately 2.58 times faster with the optimized parallel implementation compared to the original sequential execution.

To fully leverage this speedup, optimization efforts should focus on parallelizing the filtering task effectively using parallel programming techniques, such as multi-threading or SIMD (Single Instruction, Multiple Data) instructions, while ensuring efficient data distribution and synchronization among the processors.

In this example, Amdahl’s Law helps us identify the significant impact of the parallelizable task (T2) on the overall execution time. By prioritizing the optimization of the filtering task while optimizing the sequential tasks (T1 and T3) as much as possible, we can achieve substantial performance improvements in the image processing application.


Amdahl’s Law guides performance optimization by highlighting the potential speedup through parallelization and optimization. Balancing these approaches and considering scalability enables us to unlock the full potential of computing systems. By applying Amdahl’s Law, profiling, analysis, and continuous measurement, we create high-performance applications that deliver optimal results.