Performance

1 min read Last updated Sat Jun 27 2026 08:46:52 GMT+0000 (Coordinated Universal Time)

Performance Equation

CPU Time=Instruction Count×CPI×Clock Cycle Time\text{CPU Time} = \text{Instruction Count}\times \text{CPI} \times \text{Clock Cycle Time}

Here:

  • CPI - Cycles Per Instruction
    CPI=Total CPU cycles/Instruction count\text{CPI} = \text{Total CPU cycles} / \text{Instruction count}
  • IPC Instructions Per Cycle
    IPC=1/CPI\text{IPC} = 1 / \text{CPI}

Performance can be improved by:

  • Reducing instruction count (better compiler)
  • Lowering CPI (efficient microarchitecture)
  • Reducing clock cycle time (faster hardware)

Performance Measurement and Benchmarking

Performance is domain-specific. Requirements evolve over time and depend on target devices.

SPEC

Standard Performance Evaluation Corporation. Provides standardized, reproducible benchmark suites.

  • SPEC CPU: measures CPU-intensive workloads (integer and floating-point subsets)
  • Results reported as a ratio relative to a reference machine
  • Geometric mean used to combine scores across workloads

MIPS

Millions of Instructions Per Second.

MIPS=Instruction countExecution time×106\text{MIPS} = \frac{\text{Instruction count}}{\text{Execution time} \times 10^6}

Peak vs Sustained Performance

  • Peak
    Maximum theoretical throughput under ideal conditions. Rarely achieved in practice.
  • Sustained
    Actual throughput under real workloads. Reduced by memory bottlenecks, branch mispredictions, and pipeline stalls.
Was this helpful?