Article
Radiology, Nuclear Medicine & Medical Imaging
Eric E. Brost, H. Wan Chan Tseung, John A. Antolak
Summary: This study developed a GPU-based MC engine for rapid dose calculation of electron beams collimated using the conventional photon MLC. The accuracy and fast computation time of this method were validated through comparisons with other calculation methods and measurements.
Article
Computer Science, Interdisciplinary Applications
D. Kolotinskii, A. Timofeev
Summary: We present the first open-source, GPU-based code, OpenDust, for complex plasmas, which provides researchers with a user-friendly and high-performance tool for self-consistent calculations in plasma flow. The code outperforms all available codes for complex plasma simulation and can also be used for simulating larger systems of dust microparticles. OpenDust interface is written in Python, offering ease-of-use and simple installation.
COMPUTER PHYSICS COMMUNICATIONS
(2023)
Article
Chemistry, Medicinal
Regina Pikalyova, Yuliana Zabolotna, Dragos Horvath, Gilles Marcou, Alexandre Varnek
Summary: The development of DNA-encoded library (DEL) technology has brought new challenges to the analysis of chemical libraries. This study introduces the concept of chemical library space (CLS) and compares four representations obtained using generative topographic mapping. These encodings allow for effective comparison of libraries and fine-tuning of matching criteria. The proposed CLS can be used for efficient analysis and selection of chemical libraries.
JOURNAL OF CHEMICAL INFORMATION AND MODELING
(2023)
Article
Computer Science, Interdisciplinary Applications
Mingyu Yang, Ji-Hoon Kang, Ki-Ha Kim, Oh-Kyoung Kwon, Jung-Il Choi
Summary: We introduce an updated library, PaScaL_TDMA 2.0, which is capable of exploiting multi-GPU environments. The library extends its functionality to include GPU support and minimizes CPU-GPU data transfer by utilizing device-resident memory while retaining the original CPU-based capabilities. The library employs pipeline copying with shared memory for low-latency memory access and incorporates CUDA-aware MPI for efficient multi-GPU communication. Our GPU implementation demonstrated outstanding computational performance compared to the original CPU implementation while consuming much less energy.
COMPUTER PHYSICS COMMUNICATIONS
(2023)
Article
Computer Science, Interdisciplinary Applications
Zhi Li, Daniel Caviedes-Voullieme, Ilhan Oezgen-Xian, Simin Jiang, Na Zheng
Summary: The optimal strategy for solving the Richards equation numerically depends on the specific problem, particularly when using GPUs. This study investigates the parallel performance of four numerical schemes on both CPUs and GPUs. The results show that the scaling of Richards solvers on GPUs is influenced by various factors. Compared to CPUs, parallel simulations on GPUs exhibit significant variation in scaling across different code sections, with poorly-scaled components potentially impacting overall performance. Nonetheless, using GPUs can greatly enhance computational speed, especially for large-scale problems.
ENVIRONMENTAL MODELLING & SOFTWARE
(2024)
Article
Computer Science, Theory & Methods
Cody J. Balos, David J. Gardner, Carol S. Woodward, Daniel R. Reynolds
Summary: This paper discusses the implementation of GPU-accelerated time integration in scientific applications, introducing new features and presenting performance results on supercomputers, showing significant speedups when using both NVIDIA and AMD GPUs.
PARALLEL COMPUTING
(2021)
Article
Computer Science, Interdisciplinary Applications
Kohei Fujita, Takuma Yamaguchi, Yuma Kikuchi, Tsuyoshi Ichimura, Muneo Hori, Lalith Maddegedara
Summary: The cross-correlation function is widely used in time-series data analysis and improving its computation speed is crucial due to the increasing availability of massive observational data. This research focuses on optimizing the TensorFloat-32 Tensor Core operations in NVIDIA GPUs to accelerate the computation of the cross-correlation function. The proposed method significantly improves the performance of the matrix-matrix product baseline and achieves high performance in seismic interferometry using actual data.
JOURNAL OF COMPUTATIONAL SCIENCE
(2023)
Article
Engineering, Biomedical
Zhao Peng, Yu Lu, Yao Xu, Yongzhe Li, Bo Cheng, Ming Ni, Zhi Chen, Xi Pei, Qiang Xie, Shicun Wang, X. George Xu
Summary: This paper presents the development and validation of a GPU-accelerated Monte Carlo dose computing module for organ dose calculations in nuclear medicine PET/CT examinations. The results show excellent agreement with a well-tested MC code and demonstrate faster computation speed compared to the existing software.
PHYSICS IN MEDICINE AND BIOLOGY
(2022)
Article
Environmental Sciences
Xinyu Zhao, Peng Liu, Bingnan Wang, Yaqiu Jin
Summary: This research investigates the signal processing flow of passive bistatic radar and proposes a parallel signal processing scheme based on GPU architecture. The proposed scheme utilizes high-computing-power GPU as the hardware platform and CUDA as the software platform and optimizes the algorithms. Experimental results show that this scheme significantly enhances the signal processing rate for passive bistatic radar.
Article
Chemistry, Physical
Nicholas J. Browning, Felix A. Faber, O. Anatole von Lilienfeld
Summary: QML-Lightning is a PyTorch package that contains GPU-accelerated approximate kernel models, able to generate trained models in a short amount of time and provide competitive energy and force predictions. Using modern GPU hardware, learning curves for energy and force, as well as timings, were reported as numerical evidence for select legacy benchmarks from atomistic simulation.
JOURNAL OF CHEMICAL PHYSICS
(2022)
Article
Multidisciplinary Sciences
Sina Moayed Baharlou, Saeed Hemayat, Kimani C. Toussaint Jr, Abdoulaye Ndao
Summary: Metadevices are of great interest and importance in nano-scale light control and manipulation. This article introduces a framework called ParallelGDS for the generation of large metasurfaces, which significantly improves speed and reduces memory requirements compared to existing methods. This framework has the potential to revolutionize metadevice technology and enable efficient applications with large metasurfaces.
ADVANCED THEORY AND SIMULATIONS
(2023)
Article
Energy & Fuels
Zhang Chen, Jun Liu, Xinglei Liu
Summary: This study presents a novel component-oriented modeling method for DHN in IEHS, focusing on pipelines, pressure sources, and junctions. Formulas of fundamental physical processes are derived based on variable thermodynamic state of the fluid rather than predetermined constants. The proposed GPU-based parallel algorithm has achieved over 3 times performance boost compared to single CPU computing, showing the versatility and practicality of the model.
Article
Computer Science, Artificial Intelligence
Rory Mitchell, Eibe Frank, Geoffrey Holmes
Summary: SHapley Additive exPlanation (SHAP) values provide a game theoretic interpretation of machine learning model predictions. While calculating SHAP values for large decision tree ensembles is challenging, this work presents GPUTreeShap, an algorithm suitable for massively parallel computation on graphics processing units. The proposed algorithm achieves significant speedups over CPU implementations.
PEERJ COMPUTER SCIENCE
(2022)
Article
Computer Science, Hardware & Architecture
Yijie Guo, Lu Lu, Songxiang Zhu
Summary: This research proposed an accelerated solution on GPU for convolutional neural network (CNN). By optimizing data layout, grid division, block division methods, and using matrix cores and operator fusion, a pipeline algorithm was implemented. The evaluation showed that our approach achieved a speedup of 1.41x on MI210, reaching 74% of the peak value of single precision calculations.
JOURNAL OF SUPERCOMPUTING
(2023)
Article
Biochemistry & Molecular Biology
Stephen J. Trudeau, Howook Hwang, Deepika Mathur, Kamrun Begum, Donald Petrey, Diana Murray, Barry Honig
Summary: We introduce the PrePCI database that predicts interactions between 6.8 million chemical compounds and 19,797 human proteins, resulting in over 5 billion predicted interactions. The database utilizes a proteome-wide collection of structural models based on traditional techniques and the AlphaFold Protein Structure Database. Metrics based on sequence and structural similarity are established between template proteins and query proteins, and machine learning is used to derive the likelihood ratios. Chemical similarity analysis identifies other small molecules that may bind to the query protein. The database can be queried using protein or compound identifiers to obtain predicted binding information.