I'm a C programmer, not a C++ one, but at least in C, you sometimes need volatile to have a correct program.

Indeed.

The underling effect of volatile is that the compiler turns off certain optimisations to do with instruction reordering and aggressive caching in registers without stores.

However in most places that a typical non-firmware, non-driver programmer would use volatile they really should be using a memory fence (acquire, release or atomic store).