Multi-GPU Accelerated Admittance Method for High-Resolution Human Exposure Evaluation.

Area of Science:

• Computational electromagnetics
• Bioelectromagnetics
• High-performance computing

Background:

• Accurate simulation of induced electric fields in human bodies exposed to external magnetic fields is crucial for safety assessments.
• Existing methods often struggle with the computational demands of high-resolution anatomical models.

Purpose of the Study:

• To develop and present a multi-graphics processing unit (GPU) accelerated admittance method solver.
• To efficiently solve for induced electric fields in high-resolution human anatomical models under low-frequency magnetic field exposure.

Main Methods:

• Discretization of anatomical models into a 3D network of admittances.
• Application of the conjugate orthogonal conjugate gradient (COCG) iterative algorithm for solving complex linear systems.
• Acceleration of the COCG algorithm using multiple NVIDIA GPUs with asynchronous concurrent execution to overlap computation and data transfer.

Main Results:

• The COCG algorithm reduces solving time by 3.5x and storage by 40% compared to the biconjugate gradient stabilized method.
• The GPU implementation achieves a 90x speedup over a CPU implementation using four NVIDIA Tesla K20c cards versus eight CPU cores.
• Nearly linear acceleration is achieved with increasing numbers of GPU cards due to hidden communication overhead.

Conclusions:

• The implemented solver efficiently handles large-scale problems, solving whole-body models at 1-mm resolution in minutes.
• The high computational efficiency makes it practical for investigating numerous human exposure scenarios with high-resolution models, meeting international dosimetry guidelines.