Skip to question
asic.fyi

Hardware interview practice

Calculate CPI with branch and cache penalties

MediumComputer ArchitectureSystemVerilog

A five-stage pipeline completes 100 instructions. Base time is N+4 cycles. Ten mispredictions add two cycles each, five cache misses add eight cycles each, and penalties do not overlap. Write the total-cycle and CPI calculation.

Starting point

Question code

N = 100 instructions
base cycles = N + 4
10 branch misses × 2 cycles
5 cache misses × 8 cycles
no penalty overlap
Reviewed example

Work through one case

Input
Case 1: Base contribution
Case 2: Penalty contribution
Case 3: Final result
Expected output
Case 1: 100 + 4 = 104 cycles.
Case 2: 10×2 + 5×8 = 60 cycles.
Case 3: 164 cycles and CPI 1.64.

The shown result follows by applying this rule: Pipeline fill and both penalty classes are counted once. The cases also demonstrate this requirement: Report CPI = 164 / 100 = 1.64 and state the no-overlap assumption.

What to cover

Requirements

  1. Compute base time as 104 cycles.
  2. Add 20 branch-penalty cycles and 40 cache-penalty cycles.
  3. Report total cycles = 164.
  4. Report CPI = 164 / 100 = 1.64 and state the no-overlap assumption.
Continue practicing
Computer ArchitectureRecognize write-back cache behaviorComputer ArchitecturePartition a cache addressComputer ArchitectureStress a page-walk and cache race