Efficient single-precision simulations of nematohydrodynamics
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866908949950758912 |
|---|---|
| author | Amaral, Guilherme N. C. Sedahmed, Mahmoud da Gama, Margarida M. Telo Coelho, Rodrigo C. V. |
| author_facet | Amaral, Guilherme N. C. Sedahmed, Mahmoud da Gama, Margarida M. Telo Coelho, Rodrigo C. V. |
| contents | Simulations of nematohydrodynamics on graphics processing units (GPUs) are typically performed using double precision, which ensures accuracy but significantly increases computational cost. However, consumer-grade GPUs are optimized for single-precision calculations, making double-precision simulations inefficient on widely available hardware. In this work, we demonstrate that single-precision simulations can achieve the same accuracy as double-precision methods while delivering a 27-fold increase in computational speed. To achieve this, we introduce two key improvements: (i) the shifted distribution function in the lattice Boltzmann method, which mitigates precision loss at low velocities, and (ii) the use of larger time steps in the finite-difference solver, which reduces numerical errors and improves overall accuracy. We find that, unlike in double precision, accuracy in single-precision simulations follows a non-monotonic trend with respect to the finite-difference time step, revealing an optimal regime for precise computations. To illustrate the effectiveness of our approach, we simulate the dynamics of single and multiple skyrmionic tubes in Poiseuille flow. Our results confirm that optimized single-precision simulations enable fast and accurate modeling of complex nematohydrodynamic systems, making large-scale simulations feasible on standard gaming GPUs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_06610 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Efficient single-precision simulations of nematohydrodynamics Amaral, Guilherme N. C. Sedahmed, Mahmoud da Gama, Margarida M. Telo Coelho, Rodrigo C. V. Fluid Dynamics Soft Condensed Matter Computational Physics Simulations of nematohydrodynamics on graphics processing units (GPUs) are typically performed using double precision, which ensures accuracy but significantly increases computational cost. However, consumer-grade GPUs are optimized for single-precision calculations, making double-precision simulations inefficient on widely available hardware. In this work, we demonstrate that single-precision simulations can achieve the same accuracy as double-precision methods while delivering a 27-fold increase in computational speed. To achieve this, we introduce two key improvements: (i) the shifted distribution function in the lattice Boltzmann method, which mitigates precision loss at low velocities, and (ii) the use of larger time steps in the finite-difference solver, which reduces numerical errors and improves overall accuracy. We find that, unlike in double precision, accuracy in single-precision simulations follows a non-monotonic trend with respect to the finite-difference time step, revealing an optimal regime for precise computations. To illustrate the effectiveness of our approach, we simulate the dynamics of single and multiple skyrmionic tubes in Poiseuille flow. Our results confirm that optimized single-precision simulations enable fast and accurate modeling of complex nematohydrodynamic systems, making large-scale simulations feasible on standard gaming GPUs. |
| title | Efficient single-precision simulations of nematohydrodynamics |
| topic | Fluid Dynamics Soft Condensed Matter Computational Physics |
| url | https://arxiv.org/abs/2503.06610 |