GPU cluster project
St1
The Software Technology project "Exploiting massive parallelism of a GPU-based compute cluster" is organized by the "Parallel Architecture Research Group" at the faculty of Electrical Engineering. An overview of the project can be found in this PDF document.

Post-master trainees have worked in groups for several months, each group working on a unique application. They focused on the following topics:
  • Application profiling and analysis
  • Multi-core programming using OpenMP
  • Programming with vector instructions (SSE)
  • GPU programming (in CUDA)
    • Mapping of kernels to a single GPU
    • Distributing the workload towards multiple GPUs
  • Cluster programming (using MPI)
    • Distributing the workload over multiple machines (CPUs and GPUs)
    • Using load-balancing and dynamic scheduling
  • Evaluating GPU and cluster computing
  • Creating performance models for multi-GPU and cluster computing
The results of their work are presented in the form of a 4-page paper and/or a webpage. The results per group are found on this page.
GPU cluster project 2011
Contact: Cedric Nugteren & Gert-Jan van den Braak
Application Webpage Paper Contact
Neural network Link to webpageNeural Network Simulation: The recognition application Maurice Peemen
Biomedical imaging Link to webpageOptimizing a Biomedical Imaging Orientation Score Framework Cedric Nugteren
Neural network training Link to webpageEfficient mapping of the training of Convolutional Neural Networks to a CUDA-based cluster Maurice Peemen
Scene reconstruction Link to webpageOptimization of a 3D Scene Reconstruction Application Zhenyu Ye
Satellite image projection Link to webpageOptimizing the Proba-V mapping algorithm implementation for a heterogenous (CPU + GPU) environment Gert-Jan van den Braak
GPU cluster project 2012
Contact: Gert-Jan van den Braak & Maurice Peemen

In 2012 two groups of five students have worked on mapping the SIFT object recognition application to the GPU cluster. With the SIFT algorithm objects can be found in images or videos independent of their scale, orientation or illumiation. Even partially covered objects can be found.

The goal of this project is to find a coffee mug between a pile of clutter on a desk in some office. An image of the coffee mug (the object) is shown on the right. The video of the desk in the office is shown below, next to the output video with the coffee mug highlighted.

Input video
Output video

The results of the two groups can be found on their webpages: