My grading—at least when I do it myself—is intended to be a bit bimodal. Bimodal at the level of each problem or problem part. Specifically, I try to give more than half the possible points to any answer that is basically correct; and I try to give fewer than half the possible points for any answer that is basically wrong. I have observed that this translates to less partial credit than many students are used to. As a consequence of the above, you are likely to get a better score getting half the problems fully correct than all the problems sort-of-correct (whatever sort-of-correct might mean in a technical domain, which I do not know).
In the "extreme" version of this grading methodology, your score is simply the number of problems that you fully and correctly solve. I like this idea, in principle, and would love to grade this way. But I have never found it to be feasible to really implement. I do approximate it more than many professors.
I grade this way because I like things to be right and I think you should know when you don't know something. If you can't solve a problem, that doesn't bother me at all (after all, I can't solve most problems I think about, either). But certainly you should know that you aren't solving the problem. Maybe, in time, you'll get it. When you do, there should be a moment when you realize: I get it! Perhaps what make our species not just ignorant but really breathtakingly so is that not only do we know so little, but most of us haven't the slightest idea just how little we know. How can we be humble and curious when we don't even know if we're right or wrong in a domain as simple and detached from the real world as a mathematically abstracted technical problem!?
Phil Rogaway