After the short introduction of the Wigner Datacenter and the ELKH cloud, we present case studies of scientific and engineering research calculations that relied on and utilized the GPU resources in the Wigner Datacenter. We shortly present also the planned expansion of GPU capacities and related services of the ELKH cloud.

Hastlayer (https://hastlayer.com/) by Lombiq Technologies is a .NET software developer-focused, easy to use high-level synthesis tool with the aim of accelerating applications. It converts standard .NET Common Intermediate Language (CIL) bytecode into equivalent Very High Speed Integrated Circuit Hardware Description Language (VHDL) constructs which can be implemented in hardware using FPGAs. One such available target FPGA platform is the Xilinx Alveo Data Center accelerator card family which will be soon deployed in the major cloud-based data centers. This talk presents an overview of how Hastlayer integrated with the Xilinx framework.

In the talk, we try to go around the question of understanding and propose a mathematical definition for it. We demonstrate that with this definition all AI tasks can be solved easily. We also discuss the measure of relevance of the individual concepts which is crucial to make lossy compression.

Biodiversity assessment is one of the most important approaches to follow the anthropogenic-induced changes in nature. For nature conservation, it is essential to obtain information about the state of ecological systems as quickly as possible to take action. Passive acoustic techniques provide a large number of sound recordings from natural landscapes; however, the processing of these recordings can be challenging. Transforming the acoustic information into the spectrographic images and using automatic image classification methods can be a feasible approach to identify species and their abundance on the recordings. We present our latest results in the identification of bats, shrews, and birds by their vocalizations using deep convolutional neural networks; and we present how these techniques can be used in ecological research and nature conservation.

First-order phase transitions characterized by the coexistence of phases are commonly observed in the surrounding world, e.g. in the freezing of water. Continuous – second-order – phase transitions also exist in classical physics, e.g. the transition between ferro- and paramagnetism at the Curie temperature. Whereas the latter class has seen straightforward generalizations to quantum systems for decades, the notion of a first-order quantum phase transition remains to be elucidated.

Bistability in certain small quantum systems has been identified as a signature of first-order quantum phase transitions, but this is problematic: a randomly switching telegraph signal between two well-resolved attractors can also be observed in quantum dynamics distinct from phase transitions. For example, the famous electron-shelving scheme – used in atomic clocks or for qubit measurement in ion-trap quantum computers – produces a similar signal without any connection to phase transitions.

There is a missing element to support the interpretation of bistability as a first-order quantum phase transition: it must be shown that bistability is only a finite-size effect, and there exists an idealized thermodynamic limit, where temporal bistability is replaced by hysteresis. This idealized thermodynamic limit can be introduced such that the physical system remains a small quantum system with a few degrees of freedom, that is, the passage to the thermodynamic limit does not involve a quantum-to-classical transition. In this talk, I present a prototype of this procedure by constructing a finite-size scaling [1] for the recently-observed photon-blockade-breakdown effect [2,3] to justify its classification as a first-order dissipative quantum phase transition.

The work involves heavy numerics, performed on a virtual cluster defined within the Wigner Cloud. The applied software was C++QED: a versatile open-source C++ application-programming framework for simulating open quantum dynamics [4-7], developed by our group. The framework uses an adaptive-timestep version of the Monte Carlo wave-function method [8], which can be readily parallelized. In the talk, I will also sketch the computer-physics aspects of the work.

[1] Vukics A, Dombi A, Fink J M and Domokos P 2019 Quantum 3 150 URL https://doi.org/10.22331/q-2019-06-03-150
[2] Fink J M, Dombi A, Vukics A, Wallraff A and Domokos P 2017 Phys. Rev. X 7(1) 011012 URL https://doi.org/10.1103/PhysRevX.7.011012.
[3] Carmichael H J 2015 Phys. Rev. X 5(3) 031028 URL https://doi.org/10.1103/PhysRevX.5.031028.
[4] http://cppqed.sf.net
[5] Vukics A and Ritsch H 2007 Eur. Phys. J. D 44 585-599
[6] Vukics A 2012 Comp. Phys. Comm. 183 1381-1396
[7] Sandner R and Vukics A 2014 Comp. Phys. Comm. 185 2380-2382
[8] Kornyik M and Vukics A 2019 Comp. Phys. Comm. 238 88-101

CERN, the European Organization for Nuclear Research, operates the largest particle accelerator in the world and is at the core of international programs on fundamental physics research. CERN has a long tradition of collaboration and excellence in research. Quantum Technology has seen incredible growth in the past few years in all their aspects of computing, sensing, communications, and theory. After a couple of years of pilot investigations, CERN has announced in June 2020 the creation of its own Quantum Technology Initiative to explore the potential of these technologies in High-Energy Physics research and contribute to their further development. This talk highlights the main objectives and current activities.

The hypothesis, that cortical dynamics operate near criticality also suggests, that it exhibits universal critical exponents which mark the Kuramoto equation, a fundamental model for synchronization, as a prime candidate for an underlying universal model. Here, we determined the synchronization behavior of this model by solving it numerically on a large, weighted human connectome network, containing 804092 nodes, in an assumed homeostatic state. Since this graph has a topological dimension d=3.05, a real synchronization phase transition is not possible in the thermodynamic limit, still, we could locate a transition between partially synchronized and desynchronized states. At this crossover point, we observe power-law tailed synchronization durations, with \tau_t=1.2, away from the experimental values for the brain.

For comparison, on a large two-dimensional lattice, having additional random, long-range links, we obtain a mean-field value:

\tau_t = 1.6(1).

However, below the transition of the connectome we found global coupling control-parameter dependent exponents 1 < \tau_t < 2,  overlapping with the range of human brain experiments. We also studied the effects of random flipping of a small portion of link weights, mimicking a network with inhibitory interactions, and the effect of annealed noise. We found similar results as before, the control-parameter dependent exponent suggests extended dynamical criticality below the transition point.

Scientific Reports 9 (2019),  to appear in J. Neurocomputing
arXiv:1912.0601819621

Algorithmic Differentiation (AD) is a tool to evaluate derivatives of the function defined by a computer program with respect to some input variables. These derivatives can be used for various optimization tasks, for example in CFD applications for design optimization or real-time risk management in finance. AD has shown to be an essential tool to get sensitivity information in such applications due to the high computational cost of calculating finite differences. However writing an AD code for a reasonably big application is a tiresome and error-prone task, which makes it infeasible to hand-write for real-life problems. Especially to maintain an efficient parallel implementation and update it for fast-changing new hardware. AD handles a computer program as a composite function, and on this composite function, AD applies the chain rule. A common strategy is to use the elementary operations as the basic functions since and apply the chain rule on their derivatives. Over the years, there have been many tools developed to ease the cost of writing AD codes using high-level descriptions of the application or in the case of C++ with templates and operator overloads.

The reverse (adjoint) mode AD evaluates the chain rule from the right-hand side. It evaluates the derivatives of the last computational step with respect to its' input than it computes the derivatives of the previous step and so on. To evaluate the derivatives of the intermediate steps, we need the intermediate state of all variables. Thus adjoint mode AD requires storing these values and the order in which the elementary functions were performed in a data structure usually referred to as the tape. The size of the tape significantly increases the memory usage of the application.

The OPS (Oxford Parallel library for Structured mesh solvers) has shown good performance and scaling on a wide range of HPC architectures. OPS takes a high-level description embedded in C++ and generates parallel implementations targeting multiple HPC architectures. Our work aims to exploit the benefits of OPS to provide performance portable adjoint implementations for future structured mesh stencil applications using OPS with minimal modifications. In OPS already generates the implementation of all computational steps. Hence OPS can build a loop-grained DAG for the applications and create automatic checkpoints of datasets, meanwhile generating efficient parallel implementations for the adjoint loops as well.

We extended OPS with an AD aware API to generate parallel implementations of adjoint computations targeting both CPU and GPU architectures and show preliminary performance results on GPU architectures.

The IEEE-754 standard defines the behavior of floating-point representation. It is correct but comes with roundings which cause non-associativity of the operations. The order of calculations, usually relaxed in a parallel environment, affects the results. In an unstructured mesh application, the order of executing a kernel function on a set of elements usually does not create the wrong output, but there can be some regulatory requirements that request bitwise reproducible answers. There are some other approaches to solve this issue. One of these is the ReproBLAS project's binned representations. Their method causes a 5n to 9n floating-point operations overhead. Another solution can be found in Lulesh, although it is applicable only for boundary or halo values. In this work, we present a reproducible ordering for the indirect increments, combined with reproducible reductions, implemented in the OP2 domain-specific language. We evaluate our algorithms on two mini-applications:

1) Airfol, an industrially representative CFD code which is a finite volume application that solves the 2D Euler equations

2) Aero, a  2D  non-linear steady potential flow simulation.

As a result, this method can guarantee full reproducibility, even if it is run on multiple GPUs with a different number of MPI processes with the price of 1.1x-4.4x slowdown.

Authors: Bálint Siklósi [1], István Zoltán Reguly [1] and Gihan R. Mudalige [2]
Affiliations:
[1]: Pázmány Péter Catholic University, Information Technology and Bionics
[2]: The University of Warwick, Department of Computer Science
Correspondence: [email protected]

Due to the focused technological progress, quantum computer prototypes are now becoming available for the first time. These are still so-called Noisy Intermediate-Scale Quantum (NISQ) devices with around 100 qubits. Nevertheless, they are able to demonstrate quantum supremacy, i.e., they can surpass the capabilities of today's classical digital computers for some well-defined (but possibly narrow) tasks. The noise in the quantum gates is, however, still limiting the size of the quantum circuits that can be executed reliably. Thus, a lot of attention and research is directed toward finding practical use-cases for which, despite the current technological constraints, these devices can outperform classical computers. In parallel with these investigations, quantum researchers and engineers are continuing to strive for more accurate quantum gates, which may eventually lead to fully fault-tolerant quantum computing. In this talk, I will present a short (and admittedly subjective) overview of these developments and the quantum computing landscape.

In recent years, both small-scale quantum computers and quantum algorithms have been significantly improved. Quantum machine learning is a particularly promising research line. Since the quantum processor prototypes accessible at the moment contain only a few dozens of noisy qubits, algorithmic research still needs high-quality classical simulation tools. In this talk, I will survey some of the most popular software environments used for such studies, and discuss the role of high-performance computing.

Inertial confinement fusion achieved with highly energetic laser pulse is a promising method to provide clean, affordable energy in the future. The state of the art of present research and development is hindered by hydrodynamic instabilities occurring at the high compression of the target fuel. Recent studies [1,2] show that a simultaneous ignition could prevent these instabilities. Combining advancements in two fields, detonations in relativistic fluid dynamics and radiative energy deposition by plasmonic nano-shells lead to a decreased initial compression of the target. The “Nanoplasmonic laser fusion” project aims to continue these recent studies with the support of experiments designed specifically to show the benefit of using nano-technology to increase energy absorption. The theoretical group of the project aims to develop the software using the particle-in-cell (PIC) the method, extending the freely (under GNU General Public License) available and downloadable from GitHub Extendable PIC Open Collaboration (EPOCH) software, which is written in Fortran90 language and can be easily run in parallel mode, based on the algorithm of [3]. Neglecting the reflectivity of the target we propose increasing the absorptivity by using plasmonic nano-shells embedded into the target. Therefore, we achieve higher ignition temperature and radiation dominated dynamics. The plan is the study of these effects with simulation and potentially predict the experimental set-ups for fusion research.

References:
[1] L. P. Csernai, N. Kroo, and I. Papp, “Radiation dominated implosion with nano-plasmonics,” Laser Part. Beams. 36 (2), 171--178 (2018).
[2] L. P. Csernai and D. D. Strottman, “Volume ignition via time-like detonation in pellet fusion,” Laser Part. Beams. 33 (2), 279--282 (2015).
[3] M. Bonitz, G. Bertsch, V. S. Filinov, and H. Ruhl. Introduction to Computational Methods in Many-Body Physics. Cambridge University Press, 2004.

Investigation of binary stars is one of the focus areas of astronomy. While a large amount of data is being constantly obtained through the newer and newer telescopes and satellite observatories, astronomers are still facing great challenges in understanding the details of the underlying physical processes. A polytropic star model was developed and placed into the gravitational field of an accompanying star to evaluate the deformation-induced through the closeness of these stars. The model takes into account the gravitational field of both stars, the centrifugal forces of the rotating stars and as such, it can be implemented in a highly parallel manner for a GPU processor. The results were validated through careful comparison to Emden solutions of polytope models and to gravitationally equipotential surfaces of the Roche model. The developed approach will be further utilized in the modeling of neutron stars.

Planet formation is an exciting topic within the field of astronomy from both observational and theoretical points of view. Despite immense progress over the last few decades, how planets are formed within the protoplanetary disks surrounding young stars is still an open question. In this presentation, I will talk about our latest findings related to new aspects of planetary system formation. We conducted numerical hydrodynamic simulations using the GPU-based code GFARGO2 and found that a protoplanetary disk can form a cascade of self-sustaining, small-scale, dusty vortices. These vortices collect large amounts of dust and offer sites for overcoming traditional barriers to dust growth. I will discuss the implications of this phenomenon in the context of planet formation and the crucial role of GPU-based techniques in achieving these results.