MIPS and MFLOPS - Performance Measures

1. MIPS - Million Instructions Per Second:
$$MIPS = \frac{\text{Instruction Count}}{Execution\ Time \times 10^6} = \frac{f_{clock}}{CPI \times 10^6}$$
where $f_{clock}$= clock frequency, $CPI$= cycles per instruction.

Measures integer/overall CPU throughput. e.g., CPU @ 2 GHz with CPI=2: $MIPS = 2000/(2)=1000$ MIPS.

Limitations of MIPS:

• Depends on instruction set - one ISA's instruction does more work than another.
• Ignores instruction complexity and mix.
• Not reliable for comparing different architectures or compilers.

2. MFLOPS - Million Floating Point Operations Per Second:
$$MFLOPS = \frac{\text{No. of Floating Point Operations}}{Execution\ Time \times 10^6}$$
Measures capability for scientific/numerical workload (add, mul, div on floats). Higher MFLOPS = better for supercomputing.

Limitations: Only counts FP ops, ignores integer, branch, I/O; depends on program (Linpack, etc) and compiler optimization.

Comparison: MIPS $\approx$ general-purpose speed, MFLOPS $\approx$ scientific speed. MFLOPS rating typically lower than MIPS on same machine because FP ops take more cycles. Both are crude; today SPEC, CPI, execution time preferred.