Under Amdahl's fixed-workload assumptions, which part keeps its duration when more cores are added?

• The serial part of the workload
• The equally shared parallel part
• Every part of the entire workload

For fixed work with an unchanged serial fraction of 0.1, what is Amdahl's ideal limiting speedup bound?

• Tenfold speedup
• Hundredfold speedup
• Unbounded speedup

Using several cores and using SIMD data lanes are the same execution arrangement.

Circle one:   True   False

One instruction applies the same addition operation to eight data elements together. Which arrangement does this describe?

• Eight cores following separate instruction streams
• SIMD data lanes sharing an addition operation
• One core executing eight scalar additions in sequence

A fixed workload is 0.1 serial. Its serial time is unchanged; its parallel part divides equally across nine cores with no overhead. What is the ideal speedup over one core?

Answer: ______________

Adding cores saves 4 ms of computation but adds 7 ms of coordination and waiting. Why does total time increase?

• The serial fraction must have fallen to zero
• The workload must have become fully independent
• The added costs exceed the computation saving

Knowing that half a workload is parallel is enough to guarantee that buying more cores reduces its actual runtime.

Circle one:   True   False

A fixed workload has an unchanged serial fraction of 0.1. The parallel part divides equally across 30 cores with no added overhead. What is the ideal speedup over one core?

• About 30, because all work divides by the core count
• About 7.69, from dividing 1 by the normalized time
• Exactly 10, because finite cores reach the limiting bound