How This Coursework Is Graded in Practice
A substantial part of the grading procedure for this coursework is not publicly documented. Important rules concerning grade quotas, exemption from the final exam, deadlines, and the CUDA component are often communicated informally and may change from year to year. This can leave students dealing with requirements that were unclear when they started the course.
This page records the grading rules that were applied in the 2025 edition. It is intended as a practical reference for understanding how the course was actually graded that year. Future students should still confirm the current rules with the course organizers.
Exemption from the Final Exam
In 2025, exemption depended on attendance, coursework grades, the submission deadline, and the overall grade distribution of the cohort.
Students first had to satisfy the minimum academic requirement: all four assessed parts — OpenMP, MPI, MPI+OpenMP, and MPI+CUDA — had to receive at least Good. Attendance was also considered, with no more than two missed lectures normally allowed. This attendance requirement has varied between years and has sometimes been removed entirely.
Meeting these conditions only established eligibility for exemption. The number of exemptions was limited, and in practice roughly the top 20% of the cohort were exempted. The effective cutoff therefore depended on the grade distribution that year. If a large number of students received Excellent, students whose components were all graded Good could still fall outside the exemption quota.
The complete submission also had to be made before the announced deadline in order to remain eligible for exemption.
Submission and Deadline
The individual assignment stages are submitted to the group instructors. The final report and source code must additionally be provided to the course organizers.
The deadline matters directly for exemption. Even if the required grades have already been obtained, the complete coursework must be submitted before the deadline to qualify.
Students who complete the coursework early also have the option of taking an early exam, described below.
The CUDA Component
The CUDA component is graded separately by the course lecturer, A. S. Kolganov.
In 2025, students could obtain their CUDA grade through in-class CUDA programming tests. There were normally two such tests, and solutions were expected to use only CUDA features that had already been covered in the course.
The grade distribution for these tests followed a fixed quota:
- the top 25% received Excellent;
- the middle 50% received Good;
- the remaining 25% received Passed.
The exact procedure used to rank students within these groups was not specified in advance and remained at the discretion of the course organizers.
Students who obtained Good or Excellent through the in-class CUDA tests were not required to complete the separate MPI+CUDA performance assignment.
The MPI+CUDA Performance Route
Students who did not obtain Good or Excellent through the in-class CUDA tests could obtain a higher CUDA grade through the MPI+CUDA assignment.
For Good, the MPI+CUDA implementation running on a single NVIDIA Tesla P100 had to be faster than the CPU implementation running on one complete Polus POWER8 node: 20 physical CPU cores / 160 hardware threads.
For Excellent, the GPU implementation had to achieve approximately a 10× speedup over the same CPU configuration.
For this particular solver and hardware combination, the latter is a target that is practically out of reach.
The Non-CUDA Parts
The OpenMP, MPI, and MPI+OpenMP parts are graded by the instructors who receive them.
The instructor may also ask the student to defend the report, explain parts of the implementation, or answer questions about the submitted code.
Early Exam
Students who had obtained at least Passed in all four coursework components could take an early sitting of the exam before the regular examination period.
The early exam used the same written-test format as the regular final exam. It provided an additional route for students who had completed the coursework before the normal exam period.
The Final Exam
Students who are not exempted take a written test in the department’s computer lab.
The coursework grades form the main basis for the final course grade. The course materials do not specify a fixed numerical weighting between the coursework average and the exam result, and the final grading procedure is determined by the course organizers.
The final grade distribution is also subject to quotas. The exact cutoffs and the number of Excellent, Good, and Passed grades are decided by the course organizers.
A Passed grade is the minimum requirement for every coursework component. If a component has not been passed before the final exam, the exam includes additional practical questions, including programming tasks, covering that component.
