How is std::atomic implemented

I'm studying the difference between mutex and atomic in C++11.

As my understanding, mutex is a kind of lock-mechanism, which is implemented based on the OS/kernel. For example, Linux offers a mechanism, which is futex. With the help of futex, we could implement mutex and semaphore. Furthermore, I've known that futex is implemented by the low level atomic operation, such as CompareAndSet.

For std::atomic, I've known that it is implemented based on the memory model, which is introduced by C++11. However, I don't know how the memory model is implemented at the low level. If it is also implemented by the atomic operation like CompareAndSet, what is the difference between std::atomic and mutex?

In a word, if std::atomic::is_lock_free gives me a false, well, I'm gonna say that std::atomic is the same with mutex. But if it gives me a true, how is it implemented at the low level?