A while ago, I wrote a post about discovering performance difference between the STM32F103 and the CS32F103. With this in mind, I decided to run the Dhrystone benchmark on all of them. A benchmark as old as I am. I have compiled a firmware such that it contains the benchmark with O0, O1, O2, O3 and Os optimisation options. As such, this test tests both the differences between the cores and the optimisation options, as I wonder if the effects of optimisation will affect the different chips differently.

Furthermore, I did a more practical test. For this test, I have connected an SSD1331 display over the SPI bus, and I draw a changing gradient, using ucglib’s DrawGradientBox function. I use the framebuffer implementation by fornellas as without a framebuffer, ucglib’s performance is dreadful. As the stm32f103cb git only 20 KiB of RAM, I use this tiny screen with a resolution of 96×64 pixels. This test has only been run at 72 MHz, as this is a practical test.

Dhrystone

For the Dhrystone benchmark, I run the microcontrollers at three speed configurations, the default bootup speed of 8 Mhz, 48 MHz and 72 MHz. For 8 MHz I use 0 wait states, at 48 MHz I use 1 wait state, and at 72 MHz I use 2 wait states, as this is the required configuration per STM32F103 documentation. For each speed, I have compiled the Dhrystone benchmark at O0, O1, O2, O3 and Os optimisation using gcc 12.2.0 and using newlib 4.3.0 as the C library.

STM32F103

APM32F103

CH32

CS32

FCM32

GD32

HK32

MH32F103