Overview
Adding processors speeds up only the part of a program that can run in parallel. Amdahl’s law shows the limit that the serial part imposes.
Enter the parallel percentage and the number of processors.
Worked examples
| Inputs | Result |
|---|---|
| Parallel portion (%)90Processors8 | Speedup: 4.71×Efficiency: 58.8% per processorMaximum possible speedup: 10×, however many processors you add |
Show 2 more examples
| Parallel portion (%)95Processors16 | Speedup: 9.14×Efficiency: 57.1% per processorMaximum possible speedup: 20×, however many processors you add |
| Parallel portion (%)50Processors64 | Speedup: 1.97×Efficiency: 3.1% per processorMaximum possible speedup: 2×, however many processors you add |
Speedup by parallel portion
With the calculator's starting values, the speedup is 4.71×. The table shows how it changes with the parallel portion.
| Parallel portion (%) | Speedup (×) |
|---|---|
| 60% | 2.11× |
| 80% | 3.33× |
| 90%starting value | 4.71× |
| 100% | 8× |
Other inputs: Processors: 8. Change them in the calculator above for your own numbers.
How it works
Speedup = 1 ÷ ((1 − p) + p ÷ n). As n grows, the speedup approaches 1 ÷ (1 − p).
The formula
| p | Parallel fraction |
| n | Number of processors |
Practical tips
- 90% parallel code on 8 cores runs 4.71× faster, and can never exceed 10×.
- Shrinking the serial part often helps more than adding cores.
- Gustafson’s law offers a more optimistic view when the problem size grows.
Frequently asked questions
Why not 8× on 8 cores?
The serial 10% still takes the same time.
Does it include overhead?
No; communication costs make real speedups lower.