Numerical Solutions of Heat and Mass Transfer with the First Kind Boundary and Initial Conditions in Capillary Porous Cylinder Using Programmable Graphics Hardware

Hira Narang, Fan Wu, Abisoye OgunniyanInternational Journal of Advanced Computer Science and Applications 2016 · DOI

Abstract

Heat and mass transfer simulation is increasingly vital across various engineering disciplines, from industrial drying processes to the design of thermal protection systems for space vehicles. However, obtaining numerical solutions for the coupled partial differential equations (PDEs) governing these processes is traditionally a time-consuming task. This paper explores the application of General-Purpose computing on Graphics Processing Units (GPGPU) using NVIDIA’s Compute Unified Device Architecture (CUDA) to accelerate these simulations. By implementing a parallelized solver for heat and mass transfer in a capillary porous cylinder under Dirichlet (first kind) boundary and initial conditions on an NVIDIA Quadro FX 4800, we achieved significant performance improvements. Our experimental results demonstrate that GPUs can accurately perform these simulations with speedups of up to 10x compared to sequential CPU execution on an Intel Xeon E5520, establishing the GPU as a highly efficient platform for high-performance engineering physics.

Introduction

The simulation of simultaneous heat and moisture transfer through porous media is a fundamental challenge in modern engineering. These processes are typically modeled as "Two-point Initial-Boundary Value Problems" (TIBVP), defined by coupled non-linear PDEs that describe how temperature ($T$) and concentration ($C$) evolve over time and space. Traditionally, these equations are solved sequentially on a Central Processing Unit (CPU). However, as the demand for higher precision increases, leading to denser simulation grids with thousands of nodal points, the computational bottleneck of sequential processing becomes a major hurdle.

This study investigates the use of the massively parallel architecture of Graphics Processing Units (GPUs) to overcome these limitations. Unlike CPUs, which are optimized for complex serial tasks, GPUs are designed to handle thousands of simple, simultaneous operations. This paper details the implementation of a CUDA-based solver and evaluates its performance and accuracy against traditional CPU methods.

Methodology

Mathematical Modeling and Numerical Discretization

The study focuses on a capillary porous cylinder. The governing equations for heat and mass transfer are established using the Boussinesq approximation. To generalize the findings, these equations are converted into non-dimensional forms. The primary goal is to solve for the temperature and concentration distributions throughout the cylinder over time.

To solve these PDEs numerically, the Finite Difference Method (FDM) is employed. The spatial domain of the cylinder is discretized into a structured grid of nodal points $(i, j)$, where $i$ represents the axial position ($z$) and $j$ represents the radial position ($r$).

Discretization of the capillary porous cylinder into a finite difference grid.

The discretization uses fixed increments $\Delta r$ and $\Delta z$. While implicit methods like Crank-Nicolson offer superior stability, the Forward Euler Method (FEM) was selected for this implementation due to its straightforward mapping to the GPU's parallel execution model.

Parallelization with CUDA

The core of the acceleration strategy involves mapping the numerical grid to the GPU's parallel architecture. Using NVIDIA’s CUDA platform, the simulation grid is decomposed into a hierarchy of threads, blocks, and grids. In this model, each nodal point in the finite difference grid is assigned to an individual GPU thread.

The CUDA memory and thread hierarchy used for parallelizing the simulation.

As shown in the figure above, CUDA organizes threads into blocks, which are further organized into a grid. This hierarchy allows the GPU to manage thousands of threads efficiently. The implementation leverages the GPU's memory hierarchy—including local, shared, and global memory—to optimize data throughput. For example, shared memory can be used to store neighbor node values that are frequently accessed by all threads within a block, significantly reducing the latency associated with global memory access.

Results and Discussion

The performance of the GPU-based solver, running on an NVIDIA Quadro FX 4800, was compared against a sequential implementation on an Intel Xeon E5520 CPU.

Performance and Scaling

The experimental results show that the GPU implementation can significantly accelerate the simulation. For large-scale grids (up to 10,000 nodal points), the GPU achieved a maximum speedup of approximately 10x.

A critical observation was the "crossover point" in performance. For small simulations with fewer than 2,500 nodal points, the CPU actually outperformed the GPU. This is attributed to the overhead costs of initializing the GPU kernels and transferring data between the system's main memory and the GPU's dedicated memory. However, as the problem size ($N$) scales, the GPU's massive parallel throughput quickly overcomes this overhead, leading to dramatic performance gains.

Accuracy and Physical Validity

Crucially, the transition to GPU computing did not sacrifice numerical precision. The results for normalized temperature and concentration generated by the GPU were nearly identical to those from the CPU. The simulation demonstrated good convergence, reaching a stable solution within 9 to 10 iterations. Physically, the results correctly modeled the expected behavior of the system: temperature was highest at the heat source and decreased toward the ambient boundaries, while concentration followed the inverse path, consistent with the physics of mass transfer in a porous medium.

Conclusion

This research demonstrates that GPUs are a powerful and reliable tool for accelerating complex heat and mass transfer simulations. By leveraging the parallel architecture of CUDA, engineers can achieve order-of-magnitude speedups, enabling more detailed and faster design cycles. While the study used hardware from the Quadro FX 4800 era, the principles remain highly relevant, and modern GPU architectures would likely yield even more significant performance improvements. Future research will explore extending these methods to more complex geometries, such as hollow cylinders, and investigating more volatile boundary conditions to further push the boundaries of GPGPU in thermal engineering.

Summary updated

Abstract

Recently, heat and mass transfer simulation is more and more important in various engineering fields. In order to analyze how heat and mass transfer in a thermal environment, heat and mass transfer simulation is needed. However, it is too much time-consuming to obtain numerical solutions to heat and mass transfer equations. Therefore, in this paper, one of acceleration techniques developed in the graphics community that exploits a graphics processing unit (GPU) is applied to the numerical solutions of heat and mass transfer equations. The nVidia Compute Unified Device Architecture (CUDA) programming model provides a straightforward means of describing inherently parallel computations. This paper improves the performance of solving heat and mass transfer equations over capillary porous cylinder with the first boundary and initial conditions numerically running on GPU. Heat and mass transfer simulation using the novel CUDA platform on nVidia Quadro FX 4800 is implemented. Our experimental results clearly show that GPU can accurately perform heat and mass transfer simulation. GPU can significantly accelerate the performance with the maximum observed speedups 10 times. Therefore, the GPU is a good approach to accelerate the heat and mass transfer simulation
BibTeX
@article{narang2016numerical,
  author = {Hira Narang and Fan Wu and Abisoye Ogunniyan},
  title = {{Numerical Solutions of Heat and Mass Transfer with the First Kind Boundary and Initial Conditions in Capillary Porous Cylinder Using Programmable Graphics Hardware}},
  journal = {International Journal of Advanced Computer Science and Applications},
  year = {2016},
  doi = {10.14569/ijacsa.2016.070607},
}
Figure