Benchmark-Performance-Analysis-of-Raspberry-Pi-Pico-Boards

This document provides Benchmark and Performance Analysis of Raspberry Pi Pico Boards

Introduction:

An analysis of experimental results comparing the execution performance of Fibonacci number calculations, as a part of benchmarking across various Raspberry Pi Pico boards is carried out in this work. The resuts focus on ARM Cortex-M0, ARM Cortex-M33 and RISC-V cores, as implemented in the RP2040 (Raspberry Pi Pico and Raspberry Pi Pico W) and RP2350 (Raspberry Pi Pico 2) microcontrollers. Experiments are carried out using the recent MicroPython firmwares loaded in appropriate boards: Ref (screen shots given below). program is developed in MicroPython using Thonny editor.

Hardware:

Raspberry Pi Pico variant boards are connected to a 20X4 LCD module using I2C interface. Connections are shown below:

Software:

Code is developed in MicroPython and stored on the device, after dumping the appropriate bootloader as described in my earlier repository:

Sample Video:

Following is the YouTube video of my sample experiment: YouTube Link:

Motivational Website(s):Ref, and also number of YouTube presentations.

Software developed in the present work, in the form of Python code is given in the repository. Experimental results obtained in the form of tables are given below. Videos corresponding to the experimental work are also given below.

Analysis and Insights:

Performance of Raspberry Pi Pico RP2040 and Raspberry Pi Pico W RP2040

The Raspberry Pi Pico outperforms the Raspberry Pi Pico W consistently. At 250MHz, Pico RP2040(ARM Cortex-M0) takes an execution time of 0.892816s, while the Pico W takes 1.137592s, making the Pico W approximately 1.27 times slower. This may be due to the additional overhead on the CPU due to the Wireless and BlueTooth interface hardware. At125 MHz, the Pico completes Fibonaccci computation in 1.785616s versus Pico W's 2.27556s, maitaining the same performance gap.

Performance of Raspberry Pi Pico 2 RP2350 (ARM Cortex-M33 and RISC-V Hazard3)

As for as Fibonacci computation is concerned, the ARM Cortex-M33 mode in Pico 2 is significantly faster than the RISC-V Hazard 3 mode: At 250MHz, Cortex M-33 executes in 0.561120s, while RISC-V Hazard3 architecture takes 1.286060s, showing that the Cortex M-33 core is 2.29 times faster. At 125MHz, Cortex M-33 core completes the computation in 1.122360s, while RISC-V Hazard3 takes 2.2572735s, maintaining the same performance difference. however, other parameters that may affect include: Compiler optimization/ Code quality and optimization/ Memory heirarchy and access patterns/ Pipeline efficiency/ Specific use case and workload. ARM Cortex M-33 features DSP Extensions for optimized mathematical calculations.

Clock Speed Scaling

Across all tested boards and modes, execution time scales proportionally with clock speed. Halving the clock frequency results in approximately twice the execution time ( factor/scale ~2). This linear relationship aligns with theoretical expectations for CPU-bound workloads.

Graphical analysis

Below is a graphical representation of the execution times for Fibonacci calculations across different Raspberry Pi Pico boards and operating frequencies:

Literature Support for Observations

ARM Cortex-M33 vs. Cortex-M0+:

Cortex-M33 features advanced DSP instructions and a deeper pipeline, providing a significant performance advantage for tasks involving intensive calculations, such as Fibonacci sequences. Ref.

ARM Cortex-M0+ Core:

ARM Cortex-M0+ cores, like those used in the RP2040, are designed for low-power and cost-sensitive applications, resulting in relatively lower computational performance compared to higher-tier ARM Cortex cores. Ref.

RISC-V Hazard3 Core:

RISC-V cores prioritize simplicity and extensibility, which can result in slower performance for computationally intensive tasks unless specifically optimized. (Ref: The RISC-V Reader: An Open Architecture Atlas By David Patterson and Waterman)

Frequency Scaling:

The linear relationship between clock frequency and execution time for CPU-bound tasks is a well-documented phenomenon in computer architecture. (Reference: "Computer Organization and Design: The Hardware/Software Interface" by Patterson & Hennessy)

Conclusion

The Raspberry Pi Pico 2 (RP2350) demonstrates a significant improvement in performance, particularly in ARM Cortex-M33 mode, making it the best performer for computationally intensive tasks. The Raspberry Pi Pico RP2040 outperforms the Pico W due to lower hardware overhead. RISC-V cores provide a promising alternative, but their performance lags behind ARM Cortex cores for workloads like Fibonacci calculations. This analysis provides a comprehensive understanding of the performance trade-offs among different Raspberry Pi Pico boards and modes.

Work in Progress:

Fibonacci computation using other modes like: Viper mode, Earthworm mode and Native mode are in progress