Performance of arm/boards/stm32f4-discovery running at high speed

[Dolu1990]

Currently, the cortex M4 speed results are from a STM32F407VGT6 running at 16 Mhz and a flash wait state of 0. If the CPU was running at its full speed, it would suffer a wait state of 5 (168 Mhz).

It would be interresting to see the performances of the microcontroller when it run with such high wait states / frequency.

Then there is also the question of what should be the base line, a M4 running bellow its maximal speed at 16 Mhz or running at full speed.

[jeremybennett]

Hi @Dolu1990

You are correct. We took the decision to run at 16MHz, so the results would not be dominated flash memory speed. This is the speed you get out-of-the-box with this particular board, so seems appropriate. We chose the particular board because it is cheap and widely available worldwide.

It is certainly fair for the RI5CY comparison, since that is a Verilator model, and not affected at all by memory wait states.

I should be very interested to see the results of running Embench at 168MHz on the STM Discovery board (IIRC that is the maximum clock speed).

[Dolu1990]

Hi @jeremybennett

It is certainly fair for the RI5CY comparison, since that is a Verilator model, and not affected at all by memory wait states.

Right, it is a good setup to get "pure" results, it make sense to me.

I should be very interested to see the results of running Embench at 168MHz on the STM Discovery board (IIRC that is the maximum clock speed).

Sure, it would give a vision of the maximal performances of the system. If i remember properly, there is also a little instruction cache which can be enabled in the STM32F4, at the cost of determinism.

[jeremybennett]

I'll leave this open for now. I've assigned it to Paolo Savini, who did the original M4 measurements, in case he has time to try some full-speed measurements.