OPTIMIZATION OF THE CALCULATION OF KINETIC EQUATIONS OF COMBUSTION PROCESSES ON GPU USING GLOBAL MEMORY AND SHARED MEMORY

Magzhan Sarsembayev, B. A. Urmashev

2026

Full text Open access: Hybrid Metadata (CSL-JSON)
DOI 10.32014/2026.2518-1726.442
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

This paper considers the problem of improving computational performance when solving kinetic equations of combustion processes on graphics processing units. Detailed chemical kinetic mechanisms used for modeling the combustion of real fuels may include hundreds of components and thousands of elementary reactions, which leads to a significant increase in computation time when traditional CPU-based methods are used. To accelerate computations, a CUDA-oriented approach is proposed, based on distributing reactions among GPU threads and optimizing access to GPU memory. The paper compares three computational variants: sequential calculation on the CPU, parallel calculation on the GPU using global memory, and parallel calculation on the GPU using shared memory. The Aramco_Mech mechanism is used as a test mechanism. Numerical experiments were carried out for mechanisms of different sizes: from 33 components and 118 reactions to 493 components and 2716 reactions. The calculations were performed on a system with an AMD Ryzen 9 5950X 16-Core Processor, 128 GB of RAM, and an NVIDIA GeForce RTX 4080 Ti graphics adapter. The results show that as the number of reactions increases, the advantage of GPU-based computations becomes more pronounced. For the mechanism containing 493 components and 2716 reactions, the CPU computation time was 1 day 4 hours 2 minutes 4 seconds 796 milliseconds, while the GPU computation time using global memory was 21 minutes 10 seconds 591 milliseconds, and using shared memory it was 10 minutes 51 seconds 585 milliseconds. The speedup was 79.43 times for global memory and 154.89 times for shared memory. The obtained results confirm the efficiency of GPU architecture and the importance of optimal memory organization when solving large systems of chemical kinetics.