Supercomputer Research Articles

Design and Implementation of GPU Pass-Through System Based on OpenStack

The paper constructs an OpenStack cloud platform on Inspur servers, including components such as Keystone, Glance, Placement, Nova, Neutron, and Dashboard. It specifically investigates the method of constructing GPU pass-through using Virtual I/O (VFIO) technology. The aim is to directly allocate graphics card resources to cloud hosts to meet high-performance computing needs. VFIO technology can directly assign physical devices (such as graphics cards) to virtual machines, avoiding performance loss and operational complexity associated with traditional virtualization. After passing through the graphics card to the cloud host via VFIO technology, it can directly control and manage the graphics card resources within the cloud host, providing better support for high-performance computing tasks. Test results show that the GPU pass-through design has a memory bandwidth of 107.5 GB/s, a pixel fill rate of 41.6 GPixel/s, a texture fill rate of 53.1 GTexel/s, a base frequency of 1326.9 MHz, and possesses a FP32 (float) floating-point performance of 2.1TFLOPS, with 72 texture units and 48 ROPs raster operations units. Moreover, in high-load computing scenarios, the pass-through graphics card exhibits superior availability and performance.

Massively Parallel Tensor Network State Algorithms on Hybrid CPU-GPU Based Architectures.

The interplay of quantum and classical simulation and the delicate divide between them is in the focus of massively parallelized tensor network state (TNS) algorithms designed for high performance computing (HPC). In this contribution, we present novel algorithmic solutions together with implementation details to extend current limits of TNS algorithms on HPC infrastructure building on state-of-the-art hardware and software technologies. Benchmark results obtained via large-scale density matrix renormalization group (DMRG) simulations on single node multiGPU NVIDIA A100 system are presented for selected strongly correlated molecular systems addressing problems on Hilbert space dimensions up to 4.17 × 1035.

A Simple, Scalable Large Deformation Solid Mechanics Implementation in the MOOSE Framework

This paper describes a large deformation solid mechanics solver implemented as part of the freely-available and open-source MOOSE finite element simulation framework. The paper documents the choices made in developing the solid mechanics framework and describes novel formulations for the gradient operator and constitutive modeling framework made to simplify implementations of different coordinate systems, stabilized gradient operators, and different constitutive model inputs and outputs. In the process, the paper describes a new formulation that casts objective integration of the Cauchy stress as a linear transformation of the small stress rate. Finally, the paper presents key implementation details and examines the parallel efficiency of the solid mechanics solver implemented in MOOSE. The implementation retains good weak scaling efficiency beyond 1,000 parallel processes. The papers includes a discussion of the factors limiting the parallel efficiency of implicit, large deformation solid mechanics codes on current high-performance computers, with the main current limitation being the scalability of the algebraic multigrid methods used to solve the linearized equilibrium equations.

The Method for Storing a Seabed Photo Map of the During Surveys Conducted by an Autonomous Underwater Vehicle

The paper introduces a novel method for creating a photographic map of the seabed using images captured by the on-board photo and video systems of autonomous underwater vehicles (AUVs) during various missions, while incorporating navigation parameters. Additionally, it presents a new approach for storing this photo map on the on-board device in a mosaic format (tiles), which significantly accelerates operational visual inspection by enabling the automatic search and recognition of underwater objects that may exceed the coverage area of a single photograph. This capability is achieved by organizing the photo map into layers with varying zoom levels. Semi-natural experiments were conducted with data from actual missions using the real underwater vehicle demonstrate the high efficiency of the proposed method and algorithm. Unlike existing methods that form photo maps after the underwater vehicle has taken pictures of the bottom using special high-performance computers, the developed method forms a photo map directly during the movement of the vehicle, using only the computing power of the on-board computer. In addition, in the event of accidents, when it is necessary to detect objects of interest on the seabed as quickly as possible, it is necessary to provide a quick visual inspection of the generated photo map. For this purpose, we have developed an algorithm for saving a photo map in the form of a mosaic, which is widely used in interactive geographic maps, such as Google Maps. This algorithm differs from existing methods in that it selectively saves data to the on-board storage device to reduce the number of read and write operations, thus ensuring the timely operation of the entire process of creating a photo map at a given frequency of photography. After the generated map has been stored as a mosaic and a high-speed connection with the vehicle has appeared, the operator can immediately view the entire generated map using a regular web browser.

Scale-up Unlearnable Examples Learning with High-performance Computing

Scale-up Unlearnable Examples Learning with High-performance Computing

Draft genome dataset of Streptomyces griseoincarnatus strain R-35 isolated from tidal pool sediments.

The marine isolate, Streptomyces griseoincarnatus strain R-35, was isolated from marine sediments collected from the Glencairn Tidal Pool, Table Mountain National Park, Cape Town, South Africa. The genomic DNA was sequenced using the Ion Torrent GeneStudio™ S5 platform, and the de novo assembly was performed using the SPAdes assembler on the Centre for High Performance Computing (CHPC) Lengau Cluster located at the CSIR, Rosebank, South Africa. The draft genome assembly consisted of 722 contigs totaling 7,625,174 base pairs and a G+C% content of 72.2 mol%. Genome completeness and genome contamination were determined as 99.12% and 0.92%, respectively. Genome annotations performed using the Rapid Annotation with Subsystem Technology (RAST) and the Bacterial and Viral Bioinformatics Resource Centre (BV-BRC) determined the presence of 7996 coding sequences (CDS), 63 transfer RNAs (tRNAs), and six ribosomal RNAs (rRNAs). A total of 2570 hypothetical proteins were assigned, and 5246 proteins were assigned to function. The phylogenomic positioning of S. griseoincarnatus strain R-35 was determined using the Type Strain Genome Server (TYGS) and was found to be related to S. griseoincarnatus JCM 4381T, with a digital DNA-DNA hybridisation (dDDH) value of 84.1%, and an OrthoANIu value of 98.22%. The CARD RGI algorithm on Proksee predicted the presence of 6,107 antimicrobial resistance (AMR) features, 27 biosynthetic gene clusters (BGCs) were predicted using antiSMASH, while 189 carbohydrate-active enzymes (CAZymes) were predicted using dbCAN3. The raw genome sequencing data has been submitted to the National Center for Biotechnology (NCBI) under the BioProject ID PRJNA1129156 (BioSample ID Accession Number: SAMN42145163; Short Read Archive (SRA) Accession: SRR29633055; https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1129156).

An efficient Radix-4 butterfly structure based on the complex binary number system and distributed arithmetic

Complex number arithmetic is pivotal in various applications, requiring the selection of an efficient multiplier for high-performance computations. Fast Fourier transform (FFT)-based multipliers are widely employed for computing complex number products, but their reliance on using dedicated multipliers and treating the real and imaginary parts as two entities significantly add to the cost and complexity of the system. Distributed arithmetic (DA) is a technique that replaces complex multiplications with a bit-level shift and addition mechanism. The complex binary number system (CBNS) utilizes binary arithmetic, which treats the real and imaginary parts as a single entity, which can simplify complex number arithmetic and computations. This paper introduces an approach integrating the CBNS with DA in a Radix-4 decimation in time FFT 8-bit and 16-bit butterfly structure. The proposed design significantly reduces arithmetic computations and eliminates dedicated multipliers, demonstrating a reduction in power consumption, area size, and lookup tables, as well as increasing overall clock performance compared to the conventional FFT architecture on Artix-7, Kintex-7, and Virtex-7 field-programmable gate array chips.

Advances in Artificial Intelligence and Computational Methods: Enhancing Modeling, Prediction, and Optimization

bstract: This area of artificial intelligence and computational methods has witnessed tremendous development, especially with the solution of highly complex problems across scientific disciplines. This review brings together the latest advancements in the integration of AI with traditional computational techniques, focusing on their applications in numerical simulations, system optimization, and predictive modeling. Key contributions include AI-augmented finite difference methods for solving partial differential equations, advancements in material and energy systems optimization, and innovative approaches to highperformance computing. Future directions emphasize scalability, interdisciplinary applications, and the integration of emerging technologies such as quantum computing and reconfigurable hardware.

Sub-nanometer depth resolution and single dopant visualization achieved by tilt-coupled multislice electron ptychography

Real-space, three-dimensional imaging of atomic structures in materials science is a critical yet challenging task. Although scanning transmission electron microscopy has achieved sub-angstrom lateral resolution through techniques like electron ptychography, depth resolution remains limited to only 2 to 3 nanometers using single-projection setups. Attaining better depth resolution often requires large sample tilt angles and numerous projections, as demonstrated in atomic electron tomography. Here, we introduce an extension of multislice electron ptychography, which couples only a few small-angle projections to improve depth resolution by more than threefold, reaching the sub-nanometer scale and potentially approaching the atomic level. This technique maintains high resolving power for both light and heavy atoms, significantly enhancing the detection of individual dopants. We experimentally demonstrate three-dimensional visualization of dilute praseodymium dopants in a brownmillerite oxide, Ca2Co2O5, along with the accompanying lattice distortions. This approach can be implemented on widely available transmission electron microscopes equipped with hybrid pixel detectors, with data processing achievable using high-performance computing systems.

Simulation and Stability Analysis of a Coupled Parachute–Payload System

High-fidelity simulations are used to study the stability of a coupled parachute–payload system in different configurations. A 8.53 m ring–slot canopy is attached to two separate International Organization for Standardization (ISO) container payloads representing a Twenty Foot Equivalent (TEU). To minimize risk and as an alternative to a relatively expensive traditional test program, a multi-phase design and evaluation program using computational tools validated for uncoupled parachute system components was completed. The interaction of the payload wake suspended at different locations and orientations below the parachute were investigated to determine stability characteristics for both subsonic and supersonic freestream conditions. The DoD High-Performance Computing Modernization Program CREATETM-AV Kestrel suite was used to perform CFD and fluid–structure interaction (FSI) simulations using both delayed detached-eddy simulations (DDES) and implicit Large Eddy Simulations (iLES). After analyzing the subsonic test cases, the simulations were used to predict the coupled system’s response to the supersonic flow field during descent from a high-altitude deployment, with specific focus on the effect of the payload wake on the parachute bow shock. The FSI simulations included structural cable element modeling but did not include aerodynamic modeling of the suspension lines or suspension harness. The simulations accurately captured the turbulent wake of the payload, its coupling to the parachute, and the shock interactions. Findings from these simulations are presented in terms of code validation, system stability, and drag performance during descent.

SurfATT: High-Performance Package for Adjoint-State Surface-Wave Travel-Time Tomography

Abstract SurfATT is a novel software package designed for ambient noise surface-wave tomography. It employs the innovative adjoint-state travel-time tomography method with the consideration of model inhomogeneity and surface topography. Key features of SurfATT include a user-friendly interface, efficient memory utilization, and high computational performance. As a result, it capably handles surface-wave tomography for large-scale datasets and models. Benchmark tests further highlight SurfATT’s remarkable performance, particularly on high-performance computing systems and Apple M1 Max chip, underscoring its potential compatibility across diverse computing environments. Its application to the western U.S. region showcases its effectiveness in generating high-resolution tomographic images and revealing spatial correlations with tectonic features. Overall, SurfATT emerges as an efficient solution for surface-wave tomography tasks, offering researchers and practitioners a powerful tool for understanding Earth’s subsurface structures.

KYAMOS Software - Mini Review on the Computer-Aided Engineering Industry

This review underscores the transformative impact of Computer-Aided Engineering (CAE) on modern engineering, emphasizing its role in advancing sustainable and efficient technological solutions. The CAE industry is further analyzed, focusing on market trends and future directions. The CAE market is projected to grow significantly in the next five years in industries like automotive, aerospace, and energy, especially with the rapid advancements in High-Performance Computing (HPC), Artificial Intelligence (AI), Internet of Things (IoT), and Digital Twin technologies that enhance real-time optimization and predictive capabilities, thus fostering innovation in sustainable product design and performance. Lastly, this review presents KYAMOS Software and its state-of-the-art CAE solutions for tackling high-demand engineering problems, mainly concerning green technologies.

Heterogeneous Packaging Technologies for Chiplet and Memory Integration

To meet the High-Performance Computing (HPC) and Artificial Intelligence (AI) market demands of ever higher performance, lower power consumption, wider memory bandwidth with reduced latency, the interconnects connecting D2D in advanced packages are getting ever smaller with tighter bump pitch. Hybrid Copper Bonding (HCB), which can provide direct Cu-Cu connection, is replacing solder-based micro bump Thermal Compressive Bonding (TCB) for die stacking, when the bump pitch shrinks down to less than 20µm. While the interconnects are getting smaller and denser, the overall package size is getting bigger, because more chiplets and High Bandwidth Memory (HBM) need to be assembled on the same package to meet the high-performance requirements. Innovative memory integration solutions, for example memory to logic die 3D stacking, photonic HBM integration, and remote optical HBM connection, are being developed to tackle the issue alternatively. This paper presents a chiplet based heterogeneous integration platform, an advanced custom HBM option, and a menu of advanced packaging offering, which are recently built and delivered by Samsung. including Integrated Stack Capacitor (ISC), Custom HBM, Re-Distribution Layer (RDL) based Fan Out Wafer Level Packaging (FOWLP), Fan Out Panel level packaging (FOPLP), interposer and Si bridge based 2.5 D package architectures, such as I-CubeS, I-CubeR, and I-CubeE, as well as TCB and HCB based 3D IC packaging.

Optimal design of experiments in the context of machine-learning inter-atomic potentials: improving the efficiency and transferability of kernel based methods

Abstract Data-driven machine learning (ML) models of atomistic interactions are often based on flexible and non-physical functions that can relate nuanced aspects of atomic arrangements to predictions of energies and forces. As a result, these potentials are only as good as the training data (usually the results of so-called ab initio simulations), and we need to ensure that we have enough information to make a model sufficiently accurate, reliable and transferable. The main challenge stems from the fact that descriptors of chemical environments are often sparse, high-dimensional objects without a well-defined continuous metric. Therefore, it is rather unlikely that any ad hoc method for selecting training examples will be indiscriminate, and it is easy to fall into the trap of confirmation bias, where the same narrow and biased sampling is used to generate training and test sets. We will show that an approach derived from classical concepts of statistical planning of experiments and optimal design can help to mitigate such problems at a relatively low computational cost. The key feature of the method we will investigate is that it allows us to assess the quality of the data without obtaining reference energies and forces—a so-called offline approach. In other words, we are focusing on an approach that is easy to implement and does not require sophisticated frameworks that involve automated access to high performance computing.

All-optical analog differential operation and information processing empowered by meta-devices

Abstract The burgeoning demand for high-performance computing, robust data processing, and rapid growth of big data necessitates the emergence of novel optical devices to efficiently execute demanding computational processes. The field of meta-devices, such as metamaterial or metasurface, has experienced unprecedented growth over the past two decades. By manipulating the amplitude, phase, polarization, and dispersion of light wavefronts in spatial, spectral, and temporal domains, viable solutions for the implementation of all-optical analog computation and information processing have been provided. In this review, we summarize the latest developments and emerging trends of computational meta-devices as innovative platforms for spatial optical analog differentiators and information processing. Based on the general concepts of spatial Fourier transform and Green’s function, we analyze the physical mechanisms of meta-devices in the application of amplitude differentiation, phase differentiation, and temporal differentiation and summarize their applications in image edge detection, image edge enhancement, and beam shaping. Finally, we explore the current challenges and potential solutions in optical analog differentiators and provide perspectives on future research directions and possible developments.

The Progress and Prospects of Data Capital for Zero-Shot Deep Brain–Computer Interfaces

The vigorous development of deep learning (DL) has been propelled by big data and high-performance computing. For brain–computer interfaces (BCIs) to benefit from DL in a reliable and scalable manner, the scale and quality of data are crucial. Special emphasis is placed on the zero-shot learning (ZSL) paradigm, which is essential for enhancing the flexibility and scalability of BCI systems. ZSL enables models to generalise from limited examples to new, unseen tasks, addressing data scarcity challenges and accelerating the development of robust, adaptable BCIs. Despite a growing number of BCI surveys in recent years, there is a notable gap in clearly presenting public data resources. This paper explores the fundamental data capital necessary for large-scale deep learning BCI (DBCI) models. Our key contributions include (1) a systematic review and comprehensive understanding of the current industrial landscape of DBCI datasets; (2) an in-depth analysis of research gaps and trends in DBCI devices, data and applications, offering insights into the progress and prospects for high-quality data foundation and developing large-scale DBCI models; (3) a focus on the paradigm shift brought by ZSL, which is pivotal for the technical potential and readiness of BCIs in the era of multimodal large AI models.

A Descriptive Study on Interconnection Networks for Parallel Computing and Algorithm Models in Parallel Computing

In parallel computing, Interconnection networks are very crucial for efficient communication among all processors within a similar system. Parallel computing has become a crucial topic in the concern of computer science and also it is revealed to be critical when researching in high performance. The evolution of computer architectures towards an improved number of nodes, where parallelism could be the approach to option for speeding up an algorithm within the last few decades. Efficient data transfer between processors is an essential component in any large scale parallel computation. Motivated by the growing interest in parallel computers, a significant amount of theoretical research has been devoted to the area of interconnection networks for parallel computers, most of it to the packet routing (or store-and-forward) model of communication. We survey some of the major developments in this field, and discuss several new alternative models of communication, In many large scale applications, communication time dominates the execution time of the whole parallel computation. Thus, the performance of a large scale parallel computer is highly correlated with the efficiency of its network and communication algorithm. The combination of processing units build a model of computation (circuits) has gained an essential place in the area of high performance computing (HPC) due to its configuration and considerable processing supremacy that is parallel, series, etc. The aim of the Presenting this paper is study on the idea of parallel computing and its programming models and also explore some theoretical and technical concepts which can be often needed to understand the Interconnection network. In particular, we show how this technology is new in assisting the field of computational physics, especially when the issue is data parallel. In the real-life example of parallel computing, there are two queues to get a ticket of anything; if two cashiers are giving tickets to 2 persons simultaneously, it helps to save time as well as reduce complexity

Performance Portrait Method: Robust Design of Predictive Integral Controller

The performance portrait method (PPM) can be characterized as a systematized digitalized version of the trial and error method—probably the most popular and very often used method of engineering work. Its digitization required the expansion of performance measures used to evaluate the step responses of dynamic systems. Based on process modeling, PPM also contributed to the classification of models describing linear and non-linear dynamic processes so that they approximate their dynamics using the smallest possible number of numerical parameters. From most bio-inspired procedures of artificial intelligence and optimization used for the design of automatic controllers, PPM is distinguished by the possibility of repeated application of once generated performance portraits (PPs). These represent information about the process obtained by evaluating the performance of setpoint and disturbance step responses for all relevant values of the determining loop parameters organized into a grid. It can be supported by the implementation of parallel calculations with optimized decomposition in the high-performance computing (HPC) cloud. The wide applicability of PPM ranges from verification of analytically calculated optimal settings achieved by various approaches to controller design, to the analysis as well as optimal and robust setting of controllers for processes where other known control design methods fail. One such situation is illustrated by an example of predictive integrating (PrI) controller design for processes with a dominant time-delayed sensor dynamics, representing a counterpart of proportional-integrating (PI) controllers, the most frequently used solutions in practice. PrI controllers can be considered as a generalization of the disturbance–response feedback—the oldest known method for the design of dead-time compensators by Reswick. In applications with dominant dead-time and loop time constants located in the feedback (sensors), as those, e.g., met in magnetoencephalography (MEG), it makes it possible to significantly improve the control performance. PPM shows that, despite the absence of effective analytical control design methods for such situations, it is possible to obtain high-quality optimal solutions for processes that require working with uncertain models specified by interval parameters, while achieving invariance to changes in uncertain parameters.

A software module to assess the metabolic potential of mutant strains of the bacterium Corynebacterium glutamicum

Technologies for the production of a range of compounds using microorganisms are becoming increas­ingly popular in industry. The creation of highly productive strains whose metabolism is aimed to the synthesis of a specific desired product is impossible without complex directed modifications of the genome using mathematical and computer modeling methods. One of the bacterial species actively used in biotechnological production is Co­rynebacterium glutamicum. There are already 5 whole-genome flux balance models for it, which can be used for me­tabolism research and optimization tasks. The paper presents fluxMicrobiotech, a software module developed at the Institute of Cytology and Genetics of the Siberian Branch of the Russian Academy of Sciences, which implements a se­ries of computational protocols designed for high-performance computer analysis of C. glutamicum whole-genome flux balance models. The tool is based on libraries from the opencobra community (https://opencobra.github.io) within the Python programming language (https://www.python.org), using the Pandas (https://pandas.pydata.org) and Escher (https://escher.readthedocs.io) libraries . It is configured to operate on a ‘file-in/file-out’ basis. The model, environmental conditions, and model constraints are specified as separate text table files, which allows one to pre­pare a series of files for each section, creating databases of available test scenarios for variations of the model. Or vice versa, allowing a single model to be tested under a series of different cultivation conditions. Post-processing tools for modeling data are set up, providing visualization of summary charts and metabolic maps.

Intelligent energy pairing scheduler (InEPS) for heterogeneous HPC clusters

In recent years, energy consumption has become a limiting factor in the evolution of high-performance computing (HPC) clusters in terms of environmental concern and maintenance cost. The computing power of these clusters is increasing, together with the demands of the workloads they execute. A key component in HPC systems is the workload manager, whose operation has a substantial impact on the performance and energy consumption of the clusters. Recent research has employed machine learning techniques to optimise the operation of this component. However, these attempts have focused on homogeneous clusters where all the cores are pooled together and considered equal, disregarding the fact that they are contained in nodes and that they can have different performances. This work presents an intelligent job scheduler based on deep reinforcement learning that focuses on reducing energy consumption of heterogeneous HPC clusters. To this aim it leverages information provided by the users as well as the power consumption specifications of the compute resources of the cluster. The scheduler is evaluated against a set of heuristic algorithms showing that it has potential to give similar results, even in the face of the extra complexity of the heterogeneous cluster.