Below is a list of open, ongoing, and completed projects/theses. If you are intested in an open project/thesis, or you alreay have a topic and looking for a supervisor, please contact me.

Open Projects/Theses

  • Implementation and evaluation of NoCs for GPU Architecture (B.Sc/M.Sc.)

    With the latest Nvidia GPU architecture, all Streaming Multiprocessors(SMs) can access the shared memory of SMs located in the same GPU Processing Cluster(GPC). Access is handled by a network on-chip that connects all the SMs of a GPC. In our group, we modeled the SM to SM communication in GPGPU-Sim using an ideal network configuration. The aim of this work is to implement different NoCs (ring bus, crossbar...), integrate them into the simulator and evaluate their performance.
    If you are interested please contact: Tim Lühnen


    • Familiarity with Linux
    • Good knowledge of C/C++
    • (Optional) Knowledge of CUDA/GPU-Architecture

  • Optimizing Dynamic Programming Algorithms for Hopper Architecture (B.Sc./M.Sc.)
  • Evalutation of different GPU simulators (B.Sc/M.Sc.)

    Currently GPGPU-Sim is one of the most popular simulators for GPU Architecture Research. GPGPU-Sim simulates Nvidias PTX Instruction Set and can run CUDA and OpenCL applications. With Vortex there is a GPGPU available that is based on the open-source RISC-V instruction set. The GPU is OpenCL compatible and the project also offers a cycle accurate simulator. The aim of this work is to compare the architectures and the performance by running different benchmarks on these simulators.
    If you are interested please contact: Tim Lühnen


    • Familiarity with Linux
    • (Optional) Knowledge of GPU architectures
    • (Optional) Knowledge of CUDA/OpenCL

  • Benchmarking Embedded GPUs for Onboard AI Space Applications (B.Sc./M.Sc.)
  • Developing Comprehensive Error Injection Models for Embedded GPUs for Space Applications (B.Sc./M.Sc.)
  • Quantifying Performance Overhead of Embedded GPUs for Space Qualification (B.Sc./M.Sc.)
  • A Quantitative Study of Space-Grade Embedded GPUs vs Off-The-Shelf Embedded GPUs for Onboard AI Processing (B.Sc./M.Sc.)
  • A Comparative Study of LSTM Versus Transformer for Sequence Modeling Problems in GPUs (B.Sc./M.Sc.)
  • A Generic Framework for Memory Access Granularity Aware Lossless Compression for Many-core Processors (B.Sc./M.Sc.)
  • Evaluating Scalability of SYCL Applications Across HPC Clusters (B.Sc./M.Sc.)

Completed Theses

  • Memory Access Granularity Aware Lossless Compression for GPUs (M. Renz)
  • An Efficient Lightweight Framework for Porting Vision Algos on Embedded SoC (A. Ashish)
  • SystemC Modeling of a Memory Paging System for Secure Elements in SoC (M. Yuan)
  • Power Modeling of Mobile GPUs using Deep Learning (M. Neu)
  • Quantitative Cache Line Re-Use Analysis for GPUs (J. Dommes, TUB)
  • Custom Hardware Accelerator for the Smith-Waterman Algorithm (L. M. Selinka, TUB)