Skip to content

Latest commit

 

History

History
134 lines (85 loc) · 6.67 KB

SoSe2025-acc-systems.md

File metadata and controls

134 lines (85 loc) · 6.67 KB

Accelerated Computing Systems ("acc-systems-seminar") -- SoSe 2025

Chair website

The seminar is organized by the Systems Research Group at TU Munich.

Course Information

  • Language: English
  • Type: Seminar Course (BSc and MSc)
  • Module: IN0014, IN2107, IN45035
  • Prerequisites:
    • we do not have any compulsory prerequisites, but we prefer students to be proficient in the basic concepts of accelerators (GPU/FPGA), parallel computing (OpenMP/CUDA/OpenCL), operating systems (particularly device drivers), virtualization technology, or equivalent background. Please check with the instructor, if you think your background is insufficient.
    • Additional preferred pre-requisites, but not compulsory courses at TUM:
      • IN2073: Cloud Computing
      • IN2125: Virtualization Techniques
      • IN2076: Advanced Computer Architecture
      • Praktikum: Systems Programming in C++
      • Praktikum: Advanced Systems Programming in C/Rust
  • TUM Online:
    • You must register for this course in TUM Online before the course starts
  • Student note:
    • Compulsory enrollment after two weeks of the matching outcome; students who fail to de-register in this period will be registered for the exam

Course Details

This course will introduce advanced topics in hardware accelerators such as GPU, FPGA, and AI-specialized accelerators (e.g., TPU) and heterogeneous computing systems. The rise of large-scale data-intensive workloads (e.g., AI applications) has been increasing demands for using such specialized hardware in modern computer systems. The students will learn about these emerging accelerators, and system designs to integrate, virtualize and control such accelerators in real-world systems. The course will cover a range of topics including:

  • Emerging accelerators (GPUs, FPGAs, TPUs...)
  • Virtualization technology for accelerators
  • Hardware-software co-design
  • Heterogeneous task scheduling
  • Security for accelerators

Objective (expected results of study and acquired competences):

The students will read, discuss, and analyze recent papers in the top conferences of this field, e.g., ASPLOS, OSDI, USENIX ATC, EuroSys, ISCA, and MICRO. Students are expected to read papers in-depth prior to class and take turns in preparing and delivering high-quality presentations of papers. Furthermore, the students will actively participate in the discussion to critically analyze the system, and advance the state-of-the-art. More concretely, the students will learn:

  • Advanced scientific topics in accelerated computer systems
  • Scientific report preparation
  • Presentation and discussion of scientific results
  • Peer-review of the scientific reports

Teaching and learning method (transfer of skills) workload for students:

We will assign papers for discussion and research study. The students will work in a team of two students. More specifically, the course will be organized in three stages:

  1. Papers presentation phase: In this phase, the students will present the research papers assigned to them. They should clearly understand and explain the context of the work, contributions, and potential future work.
  2. Research phase: In the second phase, the student will work on the identified research problem statement. The research phase should clearly motivate the research problem, and also should clearly explain the research methodology to solve the problem.
  3. Reports and peer-review: As the last phase, the students are expected to submit a report summarizing the presented paper and the research proposal. Further, we will be using the state-of-the-art reviewing system (hotcrp) for peer-review of the reports. In the peer-review process, the students are expected to give feedback covering the positive and critical aspects.

We recommend that the students devote roughly equal amounts of time for each of these phases. It is the responsibility of the presenters to manage the discussion and cover the various questions within this time. Finally, the fellow participants are expected to read the paper and engage the presenters in an active discussion.

Registration

  • Capacity: We plan to have at most 24 places in the seminar.
  • Deadline: Two weeks after the matching period, we will formally enroll you for the course. If you want to drop out, please inform us before the registration deadline.

Assessment

The course grades will be based on student presentations (40% points), a 3-4 page project report summarizing the paper and research proposal (40% points), and class participation and peer-review (20%).

Slack workspace

We will use Slack for all communication. Please enroll in our Slack workspace using your official TUM email address.

Important resources

Please carefully go over the following resources to prepare for the seminar:

Important dates

  • Preliminary meeting (online):

  • Kick-off meeting:

    • Date: April 24 (Thu), 2025, 14:00 - 15:00 CET
    • Location: 01.07.014, Seminarraum (5607.01.014)

  • Research presentation (Day 1):

    • Date: June 5 (Thu), 2025, 13:00 - 16:30 CET
    • Location: 01.07.014, Seminarraum (5607.01.014)

  • Research presentation (Day 2):

    • Date: June 12 (Thu), 2025, 13:00 - 16:30 CET
    • Location: 01.07.014, Seminarraum (5607.01.014)

  • Report due (via HotCRP): July 4 (Fri), 2025, 23:59 CET

  • Peer-review due (via HotCRP): July 18 (Fri), 2025, 23:59 CET

Paper reading list

  • The list of papers will be released on Slack in the first week.

Report submission and peer-review portal

We will manage the submission and peer-reviewing of reports via HotCRP:

Please ensure that the email addresses used for submissions are the same as those on which the invitation to sign was received.

Contact

We strongly prefer slack for all communications. For any further questions/comments, please contact the course organizer(s):