All cards. Yes.
That's what I had to learn on. One syntax error and you had to resubmit the damn cards all over again to the ops people. We 'as students' when I first was learning this stuff were under deadlines, so having to resubmit your deck was almost fatal as far as passing the course (USAF '85). It sucked.
Several hours later (depending on your class schedule, often the next day), you'd get the deck back with a printout of your results wrapped around it with a rubberband. The printout would usually be a dump of your program's core when it aborted, and sometimes would be a vague but error-riddled version of the output you intended from the program itself.
The core dump was ordinarily much larger, so the moment the routing monkey brought your results to the window, you could get that sinking feeling right away.
The routing room was in one building and the computer was in another. They had a small computer (a 1401) in the routing room which did nothing but read the submitted card decks onto tape, which would be carried over to the computer building a few times througout the day and night. Then during the next trip, the results tape would be carried back to the routing room and the outputs of the various programs run out to a high-speed printer (a 1403) on the spot. Late-night submissions were handled therefore under sort of a FedEx business model, except that the trip was only across Springfield Avenue, and not to Memphis.
Over in the computer building was another computer with its routing room nearby. This computer was a first-generation solid state machine designed and built at the University--kind of a 1962 supercomputer. In its routing room were three lightbulbs set near the ceiling in a row. They were red, yellow, and green. They were labeled "DOWN," "?," and "UP."