This is Seismic Full Waveform Inversion (FWI) is a powerful technique used to estimate subsurface parameters by analyzing seismic measurements obtained at the surface. However, due to the large volume of data, complex model sizes, and non-linear iterative procedures involved, numerical computations for FWI are often computationally intensive and time-consuming. This project addresses these challenges by implementing parallel computation techniques with Graphical Processing Units (GPUs) via CUDA to significantly accelerate the FWI process.
Note: This project is an implementation of the research paper described in this paper.
- Host code is written in C++ to manage the overall project structure and coordinate computations.
- Parallel computation codes are written in CUDA C, a language optimized for GPU processing.
The project includes a comprehensive performance evaluation:
-
Comparing CUDA C and OpenMP: The computational time and performance achieved through CUDA C and OpenMP parallel computation are compared to a serial code implementation.
-
Scaling with Model Dimensions: The project demonstrates that as model dimensions increase, the performance improvement is enhanced. It remains nearly constant after reaching a certain threshold.
-
Impressive GPU Performance Gain: In our experiments, we achieved a GPU performance boost of up to 90 times compared to the serial code, underscoring the substantial benefits of GPU acceleration.
To run this project, you will need the following:
- NVIDIA GPU with CUDA support.
- CUDA Toolkit installed (Download CUDA Toolkit).
- C/C++ Compiler (e.g.,
nvccfor CUDA code andg++for CPU code). - Git (optional, for cloning the repository).