Introducing heterogeneous HPC solutions to UK's MMM community

Lead Research Organisation: University College London
Department Name: Centre for Advanced Research Computing

Abstract

The UK is a world leader in materials and molecular modelling (MMM), which provides fundamental insights into the processes and mechanisms that underlie physical phenomena and has become an indispensable element of contemporary materials research. It is no exaggeration to state that MMM is changing how new materials-based technologies are developed, acting as a guide for experimental research, helping to speed up progress and save resources. The rapid growth of the field has created an unprecedented need for a range of supercomputers. The MMM Hub was established in 2017 centred around the provision of a mid-sized supercomputer created as a dedicated service to address an acute need for capacity (large throughput of mid-sized simulations) and to reduce the overdemand for time on the national capability supercomputer (ARCHER, then ARCHER2) that is designed for large size simulations. The MMM community (over 200 UK-based research groups) typically consume two fifths of the compute from ARCHER and ARCHER2 and the performance of key MMM software was used in their procurement.

The MMM Hub provides leadership in the field of MMM, underpinning world-class science and working closely with other supercomputer facilities in the UK, strengthening established links and creating new opportunities of collaborations between the academic community and the industry. The MMM Hub supports scientific projects across the EPRSC priority areas covering topics such as materials for energy, continuum mechanics and fluid dynamics, biophysics and soft matter, chemical biology and biological chemistry, synergy with experiments and software engineering. Scientific progress in these areas will aid the development of more efficient and cost-effective novel materials and faster and more accurate computer codes for materials modelling. Moreover, the MMM Hub engages with the broad MMM community to develop a training program tailored to the community's needs. Key discoveries are disseminated via our website, peer-reviewed publications and conference presentations.

This proposal aims to provide the UK MMM community with an appropriate and readily accessible modern heterogenous HPC. In doing so, it will enhance the capability of the MMM Hub and introduce new users and expand the community who use heterogenous HPC. This is a major step for the MMM community as current international HPC facilities and anticipated national UK HPC facilities will be based on heterogeneous hardware, i.e. include accelerators. Although there are MMM software that can run on a GPU (i.e. an accelerator card), the majority of MMM software does not and so it is vitally important to provide training and support to encourage MMM software developers to port their codes and help users of their software to exploit the available hardware. Thus, the proposal requests funds for a set of accelerators, their installation costs and funds for support and training activities to optimise the return on this investment. The MMM community use, develop and optimise a wide range of MMM software, including the most popular software that consumed the largest portion of resources of the national ARCHER service. The choice of accelerator is carefully chosen to ensure that it is suitable for running mid-sized simulations using this most popular MMM code, and which is likely to also be suitable for the porting and optimisation of other MMM software. Lessons learned from porting and optimising one MMM code will be disseminated to the MMM community to help with the porting and optimisation of other MMM codes. This proposal will ensure that our Hub continues to properly serve the UK MMM community, to enable world-leading research in MMM and to add value to many of the almost 1,000 EPSRC-funded projects in this research area, including Centres for Doctoral Training, and flagship national activities, e.g. UK Catalysis Hub, the Faraday Institute, and the Sir Henry Royce Institute for Advanced Materials.

Publications

10 25 50

publication icon
Chaparro G (2024) Simulation and Data-Driven Modeling of the Transport Properties of the Mie Fluid. in The journal of physical chemistry. B

 
Description This award delivered a GPU based expansion to the EPSRC funded Materials and Molecular Modelling Hub service, making 48 GPUs available to the materials community attached to the 23,040 core CPU based service that they already use for materials and molecular modelling.
Exploitation Route The nodes are in service and still available to the Materials community in the UK. There are multiple access mechisms for both researchers at partner institutions and the UK generally - see https://mmmhub.ac.uk/young/ for more details.
Sectors Chemicals

Digital/Communication/Information Technologies (including Software)

Manufacturing

including Industrial Biotechology

URL https://mmmhub.ac.uk/
 
Title CCDC 2251914: Experimental Crystal Structure Determination 
Description Related Article: Qi Lin, Hao Lan, Chunmiao Ma, Ryan T. Stendall, Kenneth Shankland, Rebecca A. Musgrave, Peter N. Horton, Carsten Baldauf, Hans-Jörg Hofmann, Craig P. Butts, Manuel M. Müller, Alexander J. A. Cobb|2023|Angew.Chem.,Int.Ed.||e202305326|doi:10.1002/anie.202305326 
Type Of Material Database/Collection of data 
Year Produced 2023 
Provided To Others? Yes  
URL http://www.ccdc.cam.ac.uk/services/structure_request?id=doi:10.5517/ccdc.csd.cc2fl9dr&sid=DataCite
 
Title CCDC 2252028: Experimental Crystal Structure Determination 
Description Related Article: Qi Lin, Hao Lan, Chunmiao Ma, Ryan T. Stendall, Kenneth Shankland, Rebecca A. Musgrave, Peter N. Horton, Carsten Baldauf, Hans-Jörg Hofmann, Craig P. Butts, Manuel M. Müller, Alexander J. A. Cobb|2023|Angew.Chem.,Int.Ed.||e202305326|doi:10.1002/anie.202305326 
Type Of Material Database/Collection of data 
Year Produced 2023 
Provided To Others? Yes  
URL http://www.ccdc.cam.ac.uk/services/structure_request?id=doi:10.5517/ccdc.csd.cc2flf2k&sid=DataCite
 
Title CCDC 2252029: Experimental Crystal Structure Determination 
Description Related Article: Qi Lin, Hao Lan, Chunmiao Ma, Ryan T. Stendall, Kenneth Shankland, Rebecca A. Musgrave, Peter N. Horton, Carsten Baldauf, Hans-Jörg Hofmann, Craig P. Butts, Manuel M. Müller, Alexander J. A. Cobb|2023|Angew.Chem.,Int.Ed.||e202305326|doi:10.1002/anie.202305326 
Type Of Material Database/Collection of data 
Year Produced 2023 
Provided To Others? Yes  
URL http://www.ccdc.cam.ac.uk/services/structure_request?id=doi:10.5517/ccdc.csd.cc2flf3l&sid=DataCite
 
Title Dataset and Model for "Understanding the limits to short-range order suppression in many-component disordered rock salt lithium-ion cathode materials" 
Description Data and Model For "Understanding the limits to Short-range order Suppression in Many-Component Disordered Rock Salt Lithium-ion Cathode Materials" Paper DOI: 10.1039/D3TA02088F Raw data (VASP calculations) and production model for the associated paper. - `cluster_expansions` contains the final model used for the analysis in the paper. The model can be loaded using the `icet` cluster expansion package using ``` from icet import ClusterExpansion # requires the icet python package ce = ClusterExpansion.read("cluster_expansion.ce") ``` see the icet docs for how to manipulate this object: https://icet.materialsmodeling.org/ - `raw_data` contains the data used to fit the model. The calculation data contains the raw vasp calculations in tar.gz files, and each training generation has an associated `.json` files. The files contain Pymatgen ComputedStructureEntry objects. Probably the easiest way to load these into a python script is ``` from monty serialization import loadfn # requires the monty package training_data = loadfn("calculation_data.json") ``` - element references contains VASP calculations (stored as above) for the elemental reference calculations used to determine the formation energies of the training structures 
Type Of Material Database/Collection of data 
Year Produced 2023 
Provided To Others? Yes  
URL https://zenodo.org/record/8005678
 
Description MMM Hub: HPE / NVIDIA GPU Training Day - 31 March 2022 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Postgraduate students
Results and Impact Developments at the MMM Hub saw the addition of 6 nodes of 8 x A100 GPUs to the machine at the end of March 2022. To familiarise MMM Hub users with the upgraded system, HPE, NVIDIA and the MMM Hub ran a training day on 31 March 2022.
Year(s) Of Engagement Activity 2022
URL https://mmmhub.ac.uk/young-events-and-training/