RT-CUDA: A Software Tool for CUDA Code Restructuring

Khan, Ayaz ul Hassan; Al-Mouhamed, Mayez; Al-Mulhem, Muhammed; Ahmed, Ashraf

doi:10.1007/s10766-016-0433-6

Cited by 4 publications

(1 citation statement)

References 68 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Monte Carlo simulation itself is another naturally parallel problem as it is the averaging of many random guesses, so it is a kind of analog to the Mandelbrot set generator. It is found that CUDA performs better when transferring data to and from the GPU and that CUDA's kernel execution is also consistently faster than OpenCL, despite the two implementations running nearly identical code [12,13]. Historically, various applications have been studied on shared memory multiprocessors, GPUs, and message passing systems, and their performance evaluated on these systems [17,18,19,20,25,26,27].…”

Section: Literature Surveymentioning

confidence: 99%

Optimizing Particle Systems through CUDA-Assisted Multithreading

Sibai

Potvin

Ngo

2020

Wseas Transactions on Systems and Control

View full text Add to dashboard Cite

Particle systems present challenges that have warranted and attracted large amount of attention in both usage and optimization. The use of particle systems has driven complexity of simulation to greater needs of data size and accuracy. Optimization, thus, has become a moving target for researchers to reach. Studies show that multithreading has potential to make the simulation efficient while optimizing complex and data-intensive particle systems. The CUDA (Compute Unified Device Architecture) works with programming languages such as C/C++ and Python to make multithreaded parallel programming easier. This work serves to analyze particle systems using CUDA and provide an understanding about how various parameters such as the particle count and grid size influence the simulation performance. We improve the CUDA particles demo by Nvidia using our Python scripts and study the impact of particles and grids on execution time and throughput. Experimental results indicate that a required level of performance can be achieved by varying the number of particles, the size grids, and the orientation of grids as needed.

show abstract