Grading on a curve means setting the most mediocre student to a certain letter grade, say C, and scaling everyone up or down based on that. Means the class will have some fails, some As, and mostly Bs C's and D's

Contrast that with hard number limits (Below 50 fails or above 90 is A), which might mean the entire class could get As or Fs based on the exam difficulty and how the class does.

The first one assumes the class is somewhat normally distributed as far as knowledge on that topic. The second one assumes the teachers are infallible and create the exams at the same difficulty every time. Neither of these hold all the time, so neither method is necessarily better.