Technology Category
Data Mining & Knowledge Management: Scalability, Parallel & Distributed SystemsMachine Learning: Hardware-aware MLSearch and Optimization: Sampling/Simulation-based Search
Accelerating Particle-in-Cell Monte Carlo simulations with MPI, OpenMP/OpenACC and asynchronous multi-GPU programming
🤖 AI Summary
To address the computational bottleneck in plasma simulation for nuclear fusion reactor design, this work targets the compute-intensive nature and strong cross-scale coupling inherent in Particle-in-Cell Monte Carlo (PIC-MC) methods. We propose the first cross-architecture parallel framework integrating MPI inter-node communication, OpenMP/OpenACC heterogeneous thread collaboration, and asynchronous multi-GPU pipelined scheduling. Our approach innovatively overlaps computation, communication, and data transfer via CUDA asynchronous streams, unified memory management, and an adaptive load-balancing algorithm. Evaluated on kilo-particle-scale plasma simulations, the framework achieves up to 12.8× speedup on a 128-GPU cluster, with strong scaling efficiency of 92%. This significantly reduces simulation turnaround time and delivers a scalable, high-performance computing foundation for high-fidelity fusion plasma modeling.
Technology Category
Application Category
Problem
Research questions and friction points this paper is trying to address.
Enhancing plasma simulation speed for fusion reactor design
Optimizing hybrid parallelization with MPI, OpenMP, and OpenACC
Improving multi-GPU scalability for large-scale plasma simulations
Innovation
Methods, ideas, or system contributions that make the work stand out.
Hybrid MPI with OpenMP and OpenACC parallelization
Asynchronous multi-GPU programming for scalability
Efficient data transfer via OpenMP nowait clauses
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
KTH Royal Institute of Technology | Institute of Plasma Physics of the CAS | University of Ljubljana
