CPU Performance and Measurement Techniques

CPU Performance

• CPU performance is affected by instruction count, CPI (Clock cycles per instruction), and clock cycle time
• Execution time is calculated by multiplying instruction count, CPI, and clock cycle time
• Clock rate is the inverse of clock cycle time
• MIPS (Millions of instructions per second) is a measure of CPU performance
• MIPS can be calculated by dividing instruction count by execution time multiplied by 10^6, or by dividing clock rate by CPI multiplied by 10^6

Units of Measurement

• Reliable performance measure is time
• Execution time is measured in seconds
• Instruction count is measured in number of instructions
• Clock cycles are counted per program

Example 1.3: CPU Performance

• A program requires 2 million instructions on a 400 MHz CPU
• Average CPI across instruction types is 2.24
• MIPS rate for this program is approximately 179

CISC and RISC Architectures

• CISC (Complex Instruction Set Computing) uses complex instructions that may require multiple clock cycles
• RISC (Reduced Instruction Set Computing) uses simple instructions that can be executed within one clock cycle
• CISC instructions might be divided into multiple RISC instructions

Intel x86 Architecture

• CISC processor architecture
• Instructions are backward compatible across newer processor versions
• Has over 500 instructions

ARM Architecture

• RISC-based processor architecture
• Used in embedded systems for high performance, low power consumption, small size, and low cost
• Billions of embedded systems and millions of computers are produced each year

Computer Architecture & Organisation

• Computer architecture defines aspects visible to programmers such as instruction set (RISC or CISC), data types, I/O mechanisms, memory addressing modes
• Computer organisation defines the hidden hardware details like signals, control units, and interfaces between memory and CPU

Clock Cycle Time or Clock Rate

• Clock speed of a CPU is measured in clock cycle time or period (T)
• Clock rate or clock speed is measured in frequency (f=1/T)
• A 1 MHz processor receives 1 million clock pulses per second
• Higher clock rate means that more data can be processed within a fixed amount of time under the same conditions

Processor Performance

• Performance factors are the number of clock cycles and clock cycle time
• Execution time for a program is calculated by multiplying the number of clock cycles needed by the clock cycle period
• To improve CPU performance, you can either reduce the number of clock cycles needed for a program or reduce clock cycle time (increase clock rate)

Example 1.1: CPU Performance

• Computer A with a 2 GHz clock rate takes 10 seconds to run a program
• Computer B, designed to execute the same program in 6 seconds, requires 1.2 times the clock cycles of Computer A
• To achieve this , Computer B needs a 4 GHz clock rate

Instruction Performance

• Execution time also depends on instruction count and clock cycles per instruction
• CPI (Clock cycles per instruction) is the average number of clock cycles per instruction for a program
• CPI can vary between different instructions
• Total clock cycles needed for a program are calculated by multiplying instruction count with CPI

Example 1.2: Instruction Performance

• Two computers with the same architecture execute the same program
• Computer A has a clock cycle time of 250 ps and a CPI of 2.0
• Computer B has a clock cycle time of 500 ps and a CPI of 1.2
• Computer A is faster because it takes less time to execute the program

Memory Hierarchy

• Design constraints involve memory size, speed, and cost
• Smaller, faster, and more expensive memories are prioritized, while larger, slower, and cheaper are supplementary
• Primary memories are volatile and semiconductor-based
• Secondary memories are non-volatile and external
• Memory hierarchy exists to balance these factors

Performance of Accesses & Levels of Memory

• M1 is a smaller, faster, and more expensive memory level, while M2 is larger, slower, and cheaper
• CPU accesses data directly if it's in M1, but data in M2 needs to be transferred first
• M1 contains recently accessed instructions and data for faster access
• Hit ratio (H) indicates the probability of finding data in M1
• Average access time is calculated considering hit ratio, access times for M1 and M2

Example 2.3: Accesses with 2 Levels of Memory

• M1 has 1000 bytes with a 0.01 μs access time
• M2 has 100,000 bytes with a 0.1 μs access time
• Hit ratio is 95%
• Average access time is 0.015 μs