10 CUDA

Individual Contributions

Justus Dreßler: all members contributed equally

Thorsten Kröhl: all members contributed equally

Julius Halank: all members contributed equally

Project Plan for CUDA Implementation

Introduction: This project plan aims to implement Nvidia CUDA technology within a tight four-week timeframe to accelerate our tsunami simulations. The plan outlines realistic goals, milestones, work packages, and a schedule to efficiently execute the CUDA integration process.

What is CUDA?

CUDA is NVidias programming API for their GPUs. GPU’s are highly parallelized and can therefore be used to accelerate certain tasks. Our workload is very well suited for this kind of acceleration, as we have to perform the same calculations on a large amount of data. CUDA comes with a lot of tools to help with the development process, such as a profiler, which we will use to optimize our code. We will implement the C++ version of CUDA, which is called CUDA C++.

https://docs.nvidia.com/cuda/cuda-c-programming-guide/_images/gpu-devotes-more-transistors-to-data-processing.png

CPU vs GPU architecture

https://blogs.nvidia.com/wp-content/uploads/2012/09/gorpycuda3.png

Extending C code with CUDA

Goals

  1. Acceleration: Implement CUDA to significantly accelerate tsunami simulations by a factor of at least 2 compared to multithreaded CPU processing.

    _images/performance.png

    Our completely not made up expected performance improvement graph

  2. Compatibility: Ensure CUDA integration without compromising calculation accuracy or functionality.

Milestones with work packages and estimated time

MS-1 Preparation Package (Week 1) :

Literature Review:

  • Conduct a brief literature review to gain fundamental knowledge about CUDA

  • Summarize relevant information for the team

Test Programs with CUDA Implementation:

  • Create small test programs to understand the basic concepts of CUDA

  • Document results and experiences for the team

MS-2 Framework Analysis and Modification Package (Week 2) :

Identification of Key Areas:

  • Analyze the existing Tsunami simulation framework to identify key areas for CUDA integration

  • Document identified areas and their significance

Codebase Modification and Initial Kernel Development:

  • Modify the codebase for smooth integration of CUDA parallelization

  • Initiate the development of initial CUDA kernels for identified simulation components

  • Document changes and initial kernel developments

MS-3 CUDA Implementation Package (Week 3) :

Development of CUDA Kernels:

  • Develop CUDA kernels for individual simulation components based on identified key areas

  • Document kernel development and implemented functions

Initial Tests for Performance Improvement:

  • Conduct initial tests to identify performance improvements resulting from CUDA implementation

  • Analyze test results and document observed enhancements

MS-4 Optimization and Testing Package (Week 4) :

Optimization of CUDA Code:

  • Optimize the developed CUDA code to enhance efficiency

  • Utilize profiling tools such as NVPROF for detailed analysis and optimization

Comprehensive Validation Tests:

  • Conduct comprehensive tests to validate calculation accuracy after CUDA implementation

  • Document test results and validate performance improvements compared to previous simulations