Skip to main content

NI-HPC User Conference

The NI-HPC Champions User Conference will be held virtually on the mornings of the 27th and 28th October 2021, between 9:30 AM and 1:00 PM. It will involve talks and discussion sessions. The topics will include the research carried out in the cluster and the utilities of the HPC system for software users and developers. If you have an idea for a discussion topic then please get in touch with the organiser: Luis Fernandez Menchero .

Agenda :

Wed 27th October 2021
- 9:30 - 10:00 Key note 1: Cristin Merrit. Alces Flight. "Diversity in HPC: Three perspectives on making supercomputing more equitable"
- 10:00 - 10:15 Coffee or tea break
- 10:15 - 11:15 Discussion panel 1: "Start up barriers for access to Kelvin, how easy is it to use"
- 11:15 - 11:30 Coffee or tea break
- 11:30 - 13:00 Talks
- 11:30 - 11:50 Sirintra Nakjang. School of Medicine, Dentistry and Biomedical Sciences. QUB. "Bioinformatics analysis workflow for precision oncology"
- 11:50 - 12:10 Tianyi Ding. School of Pharmacy. QUB. "Mapping Allosteric Sites of the G Protein-coupled Receptors using NAMD on Kelvin-2."
- 12:10 - 12:30 Jaroslav Hofierka. CTAMOP, QUB. "Many-body theory of positron binding to molecules: Parallel implementation using MPI"
- 12:30 - 12:50 Julia Westermayr (EPSRC). Department of Chemistry, Warwick University. "Ionization potentials, electron affinities, and photoemission spectra predicted with machine learning"

Thu 28th October 2021
- 9:30 - 11:00 Talks
- 9:30 - 9:50 James Gillanders. Astrophysics Research Centre, QUB. "Using high-performance computing with TARDIS"
- 9:50 - 10:10 Donal McCaughey. Mechanical and Manufacturing Engineering, QUB. "Numerical Analysis of Vortices in Supersonic Flow"
- 10:10 - 10:30 Volodymyr Shentsov. Hydrogen Safety. Ulster University. "Simulation of blast wave and fireball propagation in a tunnel following high-pressure hydrogen tank rupture."
- 10:30 - 10:50 JunKyu Lee. School of Electronics, Electrical Engineering and Computer Science, QUB. "Energy-efficient transprecision computing."
- 11:00 - 11:15 Coffee or tea break
- 11:15 - 12:15 Discussion panel 2: "Building towards a better HPC community"
- 12:15 - 12:30 Coffee or tea break
- 12:30 - 13:00 Key note2: Dmitrios Nikolopoulos. Computer Science, Virginia Polytechnic Institute and State University. "Programming Models for High-Performance Computing: Key Drivers and Barriers to Adoption."


To register :



  • Presenter: Cristin Merrit. Alces Flight.
    Title: "Diversity in HPC: Three perspectives on making supercomputing more equitable"
    Supercomputing is becoming more open and accessible - and with that great power comes great responsibility. How can we take this opportunity to make HPC more equitable? Speakers Cristin Merritt and Wil Mayers of Alces Flight will walk through three perspectives that are shaping the future of how HPC is utilised. From how to solve problems, to how to attract research, to becoming more inclusive - our speakers will highlight key concepts that anyone working in or around HPC can use to take their projects to the next level.


  • Presenter: Dmitrios Nikolopoulos. Computer Science, Virginia Polytechnic Institute and State University.
    Title: "Programming Models for High-Performance Computing: Key Drivers and Barriers to Adoption."
    The landscape of programming models for High-Performance Computing (HPC) is ever expanding. While significant investment in new programming models and paradigms for High-Performance Computing has been driven by the global investment in Exascale Computing, the state of practice in programming models suggests that the greatest returns in adoption and productivity come from programming models that are domain-specific and favor simplicity and ease of use over absolute peak performance. This talk will explore the trade-off between performance and productivity in programming models for HPC and argue that high-productivity models are going to be more responsive to future changes in the software and systems landscape for HPC.


  • Presenter: Sirintra Nakjang. School of Medicine, Dentistry and Biomedical Sciences. Queen's University Belfast.
    Title: "Bioinformatics analysis workflow for precision oncology"
    Advances in high-throughput sequencing (HTS) technologies and the improvement in bioinformatics analyses have revolutionized the field of precision cancer medicine. This has significantly advanced the clinical application of cancer genomics by enabling high-quality and comprehensive molecular characterization of patient tumours in a timely manner to help inform tumour diagnosis and to drive therapeutic decision in daily practice. At the PMC genomics, we have developed a reproducible and scalable bioinformatics analysis workflow to facilitate the analysis of tumour DNA sequencing data generated from HTS technology. We used the workflow management system called Snakemake in combination with containerization technology (Singularity) to ease the portability and scalability of the workflow. The analysis steps include quality check, and assessment of single nucleotide variants, insertions/deletions, gene copy-number alteration and chromosomal rearrangement. The workflow has now been optimised and run successfully on kelvin2 HPC and, so far, has processed ~50-100 tumour samples per month.


  • Presenter: Tianyi Ding. School of Pharmacy. Queen's University Belfast.
    Title: "Mapping Allosteric Sites of the G Protein-coupled Receptors using NAMD on Kelvin-2."
    Drugs with increased selectivity and potentially reduced side effects could be identified through targeting allosteric sites of G protein-coupled receptors (GPCRs) [1]. We developed a probe confined dynamic mapping methodology that helps to predict GPCR allosteric sites either at extracellular or intracellular sides, as well as at the receptor–lipid interface. The applied harmonic wall potential enhanced sampling of probe molecules in a chosen GPCR region and prevents diffusion to membrane or water in molecular dynamics (MD) simulations. The M2 muscarinic, beta2 adrenergic, and P2Y1 purinergic receptors were selected for methodology testing. The MD simulations were conducted using the NAMD2 software on Kelvin2. NAMD2 shows a higher efficiency when using the combination of GPU and CPUs rather than using a numbers of CPU cores. Our methodology provides fast and efficient prediction of key amino acid residues surrounding allosteric sites in membrane proteins and facilitates the structure-based design of allosteric modulators [2].
    [1] Keov, P.; Sexton, P. M.; Christopoulos, A. Allosteric Modulation of G Protein-Coupled Receptors: A Pharmacological Perspective. Neuropharmacology 2011.
    [2] Ciancetta, A, Gill, A.K.; Ding, T.; Karlov, D.S.; Chalhoub, G.; McCormick P.J.; Tikhonova I.G.; Probe Confined Dynamic Mapping for G Protein-Coupled Receptor Allosteric Site Prediction. ACS Cent. Sci. 2021


  • Presenter: Jaroslav Hofierka. Centre of Theoretical Atomic, Molecular, and Optical Physics (CTAMOP). Queen's University Belfast.
    Title: "Many-body theory of positron binding to molecules: Parallel implementation using MPI."
    Whilst positron binding energies have been measured for over 90 molecules, an accurate ab initio theoretical description has remained elusive. Of the molecules studied experimentally, ab initio calculations exist for only 6, and for these, standard quantum chemistry approaches have proved severely deficient, agreeing with experiment to at best 25% accuracy.
    We develop a diagrammatic many-body description of positron-molecule interactions and uncover the role of strong many-body correlations [1]. We solve the Dyson equation for the positron quasiparticle wavefunction in a Gaussian basis, constructing the positron-molecule selfenergy including the GW contribution that describes polarisation, screening and electron-hole interactions, and the ladder series of positron-electron interactions that describes the unique virtual positronium formation process (where a molecular electron temporarily tunnels to the positron). We implement the above in the massively-parallelised EXCITON+ code employing MPI and state-of-the-art numerical libraries including Scalapack. Overall, we calculate binding energies in excellent agreement with experiment.
    [1] J. Hofierka, B. Cunningham, C. M. Rawlins, C. H. Patterson and D. G. Green. 2021.


  • Presenter: Julia Westermayr. Department of Chemistry. University of Warwick.
    Title: "Ionization potentials, electron affinities, and photoemission spectra predicted with machine learning."
    High-throughput screening of molecular excited states can greatly accelerate the search for functional organic molecules with potential relevance in optoelectronics. However, experimental characterization of candidate molecules is time-consuming and accurate computational methods are severely limited by the associated high computational costs [1].
    In this talk, we will show how two interdependent machine learning models can be used enable a computationally efficient description of molecular excited states with near experimental accuracy to calculate ionization potentials, electron affinities, and photoemission spectra. The basis of the newly developed method is a physically inspired machine learning model that represents orbital energies as eigenvalues of a latent Hamiltonian matrix. The eigenvalues are then corrected with another model to obtain quasiparticle energies with the accuracy of the GW approximation [2]. The performance of the method is demonstrated by predicting molecular resonances of thousands of molecules of the spectroscopy data set "OE62"[3]. Furthermore, the ability to go beyond learned systems is showcased using unseen molecules with potential relevance to new electronic devices [2].
    [1] J. Westermayr and P. Marquetand. Chem. Rev., 121, 16, 9873-9926 (2021)
    [2] J. Westermayr and R. J. Maurer. Chem. Sci. 12, 10755-10764 (2021)
    [3] A. Stuke, C. Kunkel, D. Golze, M. Todorović, J. T. Margraf, K. Reuter, P. Rinke, and H. Oberhofer, Sci. Data. 7, 58 (2020)


  • Presenter: James Gillanders. Astrophysics Research Centre. Queen's University Belfast.
    Title: "Using high-performance computing with TARDIS."
    My PhD research focusses on trying to model one of the most extreme astrophysical explosion sites - kilonovae. These are the result of the violent merger of two neutron stars, the densest objects known in the Universe. To model these extreme explosions, we use a 1D Monte Carlo radiative transfer spectral synthesis code, TARDIS. This is an extremely powerful tool for modelling the properties of these, and similar, highly energetic explosions. It simulates how photons move through the explosion, and then makes predictions for how these explosions would look to an observer. Through comparisons between the predicted models and observations, we can deduce properties of the material ejected from these explosions. There are many free parameters present in the code, since these events are extremely complex, and contain extensive and detailed physical processes as photons interact with atoms and ions moving at high velocities. This leads to a computationally intensive modelling problem. To circumvent this issue of many free parameters, we generate grids of models, that each explore only a small subset of the ejecta properties. It is through iterative studies like this that we can converge towards values that best represent the observations. In this talk I will summarise how Kelvin-2 has dramatically improved our ability to carry out this type of study.


  • Presenter: Donal McCaughey. School of Mechanical and Aerospace Engineering. Queen's University Belfast.
    Title: "Numerical Analysis of Vortices in Supersonic Flow."
    Vortices in supersonic flow may be used to accelerate mixing in hypersonic jet engines. Turbulent flows such as these flows, span a range of length scales. This creates a highly discretised and complex problem where fluid and flow properties must be calculated across millions of cells for thousands of iterations. Using the resources available on the HPC enables higher resolution simulations and decreased computation time, allowing for more detailed analysis of the physics to achieved in a reasonable time frame. For this project, the open-source software SU2 is used as solver; SU2 can run in parallel across multiple nodes (MPI) with no additional license requirements. While SU2 can scale well across multiple nodes, the additional communications overheads have shown to be a critical factor which can slow down computation significantly. These communications overhead have shown to decrease the speed up of a factor of 3 during peak time usage of the cluster.


  • Presenter: JunKyu Lee. School of Electronics, Electrical Engineering and Computer Science.
    Title: "Energy-efficient transprecision computing."
    Linear systems appear in many applications such as signal processing, electromagnetic simulation, quantum scattering, spectrophotometry, and scientific computation. It is challenging to maintain the power budget low for large-scale linear system applications, since parallel computation obtains speedup, while requiring additional power according to the increased number of processors. In modern processors, lower precision arithmetic reduces execution time, engendering less energy consumption. Our research explores transprecision computing that utilizes various precision arithmetic for linear solvers to reduce energy consumption by obtaining speedup without increasing number of processors. Our transprecision linear solver exploits runtime information including runtime convergence and computational latencies to select the least sufficient precision dynamically. Our techniques were implemented on an Intel Xeon 2.4GHz core and brought 1.8-2.4X energy reduction to a previously proposed mixed precision iterative refinement when a double precision solution accuracy was required, and a matrix size was ranged from 4K to 32K.