Posted on March 22, 2018 at 16:28I had the same observations and it took me a while to understand why. The STM32F7 is only faster if you enable the ICTM and DCTM and place time critical sections of the code judiciously. By default the compiler will not do this, so unless you know what you are doing and how to take advantage of these additional features, the STM32F7 will appear slower than the STM32F4.
In other words, you need more fine tuning on the STM32F7 and the default compiler options can’t do this for you because they don’t know what sections of your code should be placed into ICTM and DCTM.
One book that explains some of these feature differences is 'The Designer's Guide to the Cortex-M Processor Family, Second Edition' by Trevor Martin. There is also an ST application note (AN4667, 'STM32F7 Series system architecture and performance') that goes into the details.
It would be difficult to offer general advice but a few suggestions: relocate your stack and time-critical data to DTCM and place your critical code (interrupt routines) to the ICTM.
Don't expect miraculous improvements, the STM32F4 is already a fast processor and ST have done a lot to make it as efficient as possible, but as an illustration, I have managed to achieve an improvement of about 20-30% in my single precision floating point intensive interrupt routine (at the same clock frequency).
