Using machine learning hardware to solve linear partial differential equations with finite difference methods
(2025) In International Journal of Parallel Programming 53.- Abstract
- This study explores the potential of utilizing hardware built for Machine Learning (ML) tasks as a platform for solving linear Partial Differential Equations via numerical methods. We examine the feasibility, benefits, and obstacles associated with this approach. Given an Initial Boundary Value Problem (IBVP) and a finite difference method, we directly compute stencil coefficients and assign them to the kernel of a convolution layer, a common component used in ML. The convolution layer’s output can be applied iteratively in a stencil loop to construct the solution of the IBVP. We describe this stencil loop as a TensorFlow (TF) program and use a Google Cloud instance to verify that it can target ML hardware and to profile its behavior and... (More)
- This study explores the potential of utilizing hardware built for Machine Learning (ML) tasks as a platform for solving linear Partial Differential Equations via numerical methods. We examine the feasibility, benefits, and obstacles associated with this approach. Given an Initial Boundary Value Problem (IBVP) and a finite difference method, we directly compute stencil coefficients and assign them to the kernel of a convolution layer, a common component used in ML. The convolution layer’s output can be applied iteratively in a stencil loop to construct the solution of the IBVP. We describe this stencil loop as a TensorFlow (TF) program and use a Google Cloud instance to verify that it can target ML hardware and to profile its behavior and performance. We show that such a solver can be implemented in TF, creating opportunities in exploiting the computational power of ML accelerators for numerics and simulations. Furthermore, we discover that the primary issues in such implementations are under-utilization of the hardware and its low arithmetic precision. We further identify data movement and boundary condition handling as potential future bottlenecks, underscoring the need for improvements in the TF backend to optimize such computational patterns. Addressing these challenges could pave the way for broader applications of ML hardware in numerical computing and simulations. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/02d12439-1202-4e7f-9483-cb9fe852c307
- author
- Boulasikis, Michail
LU
; Gruian, Flavius
LU
and Szász, Robert-Zoltán LU
- organization
- publishing date
- 2025-03-04
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- AI hardware, Partial differential equations, Finite difference methods
- in
- International Journal of Parallel Programming
- volume
- 53
- article number
- 15
- pages
- 22 pages
- publisher
- Springer Nature
- ISSN
- 1573-7640
- DOI
- 10.1007/s10766-025-00791-6
- project
- Employing AI Hardware for General Purpose Computing
- language
- English
- LU publication?
- yes
- id
- 02d12439-1202-4e7f-9483-cb9fe852c307
- date added to LUP
- 2025-03-07 13:43:35
- date last changed
- 2025-04-04 14:57:41
@article{02d12439-1202-4e7f-9483-cb9fe852c307, abstract = {{This study explores the potential of utilizing hardware built for Machine Learning (ML) tasks as a platform for solving linear Partial Differential Equations via numerical methods. We examine the feasibility, benefits, and obstacles associated with this approach. Given an Initial Boundary Value Problem (IBVP) and a finite difference method, we directly compute stencil coefficients and assign them to the kernel of a convolution layer, a common component used in ML. The convolution layer’s output can be applied iteratively in a stencil loop to construct the solution of the IBVP. We describe this stencil loop as a TensorFlow (TF) program and use a Google Cloud instance to verify that it can target ML hardware and to profile its behavior and performance. We show that such a solver can be implemented in TF, creating opportunities in exploiting the computational power of ML accelerators for numerics and simulations. Furthermore, we discover that the primary issues in such implementations are under-utilization of the hardware and its low arithmetic precision. We further identify data movement and boundary condition handling as potential future bottlenecks, underscoring the need for improvements in the TF backend to optimize such computational patterns. Addressing these challenges could pave the way for broader applications of ML hardware in numerical computing and simulations.}}, author = {{Boulasikis, Michail and Gruian, Flavius and Szász, Robert-Zoltán}}, issn = {{1573-7640}}, keywords = {{AI hardware; Partial differential equations; Finite difference methods}}, language = {{eng}}, month = {{03}}, publisher = {{Springer Nature}}, series = {{International Journal of Parallel Programming}}, title = {{Using machine learning hardware to solve linear partial differential equations with finite difference methods}}, url = {{http://dx.doi.org/10.1007/s10766-025-00791-6}}, doi = {{10.1007/s10766-025-00791-6}}, volume = {{53}}, year = {{2025}}, }