Skip to content
Open access

PERFORMANCE COMPARISON OF EXPLICIT EULER IMPLEMENTATIONS FOR AN ENSEMBLE OF RUCKLIDGE SYSTEM TRAJECTORIES IN MATLAB

Sep 2026 · ВИМІРЮВАЛЬНА ТА ОБЧИСЛЮВАЛЬНА ТЕХНІКА В ТЕХНОЛОГІЧНИХ ПРОЦЕСАХ · 0 citations · 5 references

Abstract

Large-scale integration of independent trajectories of dynamic system is limited not only by the cost of the recurrence step but also by the organization of parallel work, storage of intermediate states, and transfers between memory levels. This study compares explicit Euler implementations for an ensemble of Rucklidge-system trajectories under three output scenarios. A 200 × 200 parameter grid was evaluated with scalar and vectorized MATLAB code on a central processing unit (CPU), ThreadPool/parfor, gpuArray operations on a graphics processing unit (GPU), specialized Compute Unified Device Architecture (CUDA) kernels, and C++ MATLAB executable (MEX) code with Open Multi-Processing (OpenMP) and single instruction, multiple data (SIMD). Each of the 40,000 trajectories contained 307,674 Euler updates. Most configurations were measured in 15 complete repeats and summarized by the median and interquartile range (IQR). With only the final state retained in double precision, two-level CPU parallelism reduced the median time from 45.232 s (IQR 0.744 s) to 2.076 s (IQR 0.163 s), a factor of 21.78. A CUDA kernel that executed 5,000 sequential steps per launch required 2.119 s (IQR 0.0014 s) in double precision and 0.291 s (IQR 0.0014 s) in single precision. For full-state storage, CUDA strict single generated the time series in video memory in 0.598 s (IQR 0.00035 s), but sequential transfer of all blocks to host memory increased the estimated pipeline time to 25.35 s. The 18.873 s CPU parfor full-history result is a critical-path estimate based on the maximum worker time, not host-observed end-to-end runtime. The results show that implementation choice depends on the parallel axis, spatial and temporal block sizes, memory locality, arithmetic precision, and the location of downstream analysis.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.