3: Performance metrics (clock rate, CPI, execution time)

Performance Metrics: Clock Rate, CPI, and Execution Time

Performance evaluation is critical in computer architecture to compare systems and optimize designs. Three core metrics form the foundation: execution time, clock rate, and CPI (Cycles Per Instruction).

Execution Time

Execution time (or wall-clock time) is the most direct measure: the total time taken by a processor to complete a specific task or program. It is the primary indicator of performance—lower execution time means higher performance. Execution time can be divided into:

• Response time: Time from task start to completion.
• Throughput: Total work done per unit time (e.g., tasks/hour).

Clock Rate

Clock rate (clock frequency) is the speed at which a processor’s clock generator emits pulses, measured in hertz (Hz). Each pulse triggers a new hardware operation cycle. A 3 GHz processor executes $3 \times 10^9$ clock cycles per second. While a higher clock rate often suggests faster processing, it alone doesn’t determine performance—efficiency in cycle usage (CPI) is equally vital.

CPI (Cycles Per Instruction)

CPI quantifies the average number of clock cycles a processor requires to execute a single instruction. It reflects hardware efficiency and is influenced by:

• Microarchitecture design (e.g., pipelining effectiveness).
• Instruction mix complexity (e.g., floating-point operations vs. integer adds).
• Memory access delays.
  A lower CPI indicates more efficient instruction processing.

The CPU Performance Equation

These metrics interrelate via the fundamental performance equation:
$\text{Execution Time} = \frac{\text{Instruction Count} \times \text{CPI}}{\text{Clock Rate}}$

• Instruction Count: Total instructions executed (determined by program/compiler).
• CPI: Average cycles per instruction (determined by hardware/instruction mix).
• Clock Rate: Cycles per second (determined by hardware technology).

Example: A program with $10^9$ instructions, running on a 2 GHz CPU with an average CPI of 1.5:
$\text{Execution Time} = \frac{10^9 \times 1.5}{2 \times 10^9} = \frac{1.5}{2} = 0.75 \text{ seconds}$

Key Considerations

• Trade-offs: Increasing clock rate may raise power consumption and heat. Reducing CPI often requires complex hardware (e.g., deeper pipelines, caches).
• Variable CPI: Individual instructions have different CPI values (e.g., ALU ops vs. memory loads). Overall CPI is a weighted average.
• Amdahl’s Law: Optimizing one component (e.g., clock rate) has diminishing returns if other components (e.g., memory latency) dominate.

Understanding these metrics enables informed design choices and accurate performance comparisons between systems.