Poster (Other)

FZJ-2023-05116

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png MSA for Polymer Simulations: Optimizing GPU and CPU Simulations of MExMeMo

Dabah, A. (Corresponding author)FZJ* ; Herten, A.FZJ*

2023

This record in other databases:

Please use a persistent id in citations: doi:10.34734/FZJ-2023-05116

Abstract: The MExMeMo Project aims to create a digital twin for the optimization, validation, and fabrication of polymers (materials) by combining two very different simulations in time, length, and scale using Modular Supercomputing Architecture (MSA). On one hand, particle dynamic methods track individual particles or elements, offering a higher level of detail and accuracy in capturing the behavior of materials. However, this level of detail increases computational requirements, including longer simulation times, larger memory requirements, and higher processing power, making it a good candidate for GPUs (JUWELS Booster). On the other hand, continuum methods provide averaged results over a given area, sacrificing some level of detail in favor of computational efficiency and high scalability, and therefore map well to CPUs (JUWELS Cluster).JSC aims to enable such workloads using the MSA concept developed as part of the DEEP projects on heterogeneous architectures. JSC also contributes to building a coordinator framework for orchestrating and managing resources and data flow between the two coupled applications, as well as optimizing the two simulations on their respective hardware. Preliminary optimization results allowed for doubling the memory throughput of the SOMA application and improving the relevant metric (TPS, timesteps per second) up to 2.2x, resulting in a reduction in power consumption by a factor of three. In the future, the project aims to incorporate a machine-learning solution for defect detection and develop a decision framework to reduce time-to-solution/energy-to-solution. Finally, the project aims to validate simulation data via a vertical integration of simulation results with membrane fabrication.

Note: The MExMeMo Project aims to create a digital twin for the optimization, validation, and fabrication of polymers (materials) by combining two very different simulations in time, length, and scale using Modular Supercomputing Architecture (MSA). On one hand, particle dynamic methods track individual particles or elements, offering a higher level of detail and accuracy in capturing the behavior of materials. However, this level of detail increases computational requirements, including longer simulation times, larger memory requirements, and higher processing power, making it a good candidate for GPUs (JUWELS Booster). On the other hand, continuum methods provide averaged results over a given area, sacrificing some level of detail in favor of computational efficiency and high scalability, and therefore map well to CPUs (JUWELS Cluster).JSC aims to enable such workloads using the MSA concept developed as part of the DEEP projects on heterogeneous architectures. JSC also contributes to building a coordinator framework for orchestrating and managing resources and data flow between the two coupled applications, as well as optimizing the two simulations on their respective hardware. Preliminary optimization results allowed for doubling the memory throughput of the SOMA application and improving the relevant metric (TPS, timesteps per second) up to 2.2x, resulting in a reduction in power consumption by a factor of three. In the future, the project aims to incorporate a machine-learning solution for defect detection and develop a decision framework to reduce time-to-solution/energy-to-solution. Finally, the project aims to validate simulation data via a vertical integration of simulation results with membrane fabrication. MExMeMo is funded by the Bundesministerium für Bildung und Forschung (BMBF) under grant 16ME0660. BMBF receives funds from the European Union-NextGenerationEU.

---

Contributing Institute(s):

1. Jülich Supercomputing Center (JSC)

Research Program(s):

1. 5112 - Cross-Domain Algorithms, Tools, Methods Labs (ATMLs) and Research Groups (POF4-511) (POF4-511)
2. MExMeMo - Virtuelles skalenübergreifendes Design zur Teilchensimulation mittels Modularem (16ME0660) (16ME0660)
3. ATML-X-DEV - ATML Accelerating Devices (ATML-X-DEV) (ATML-X-DEV)

Appears in the scientific report 2023

---

Database coverage:

---