TY - GEN
AB - We push the boundaries of electronic structure-based ab-initio
molecular dynamics (AIMD) beyond 100 million atoms. This scale is otherwise
barely reachable with classical force-field methods or novel neural network and
machine learning potentials. We achieve this breakthrough by combining
innovations in linear-scaling AIMD, efficient and approximate sparse linear
algebra, low and mixed-precision floating-point computation on GPUs, and a
compensation scheme for the errors introduced by numerical approximations.
The core of our work is the non-orthogonalized local submatrix (NOLSM)
method, which scales very favorably to massively parallel computing systems and
translates large sparse matrix operations into highly parallel, dense matrix
operations that are ideally suited to hardware accelerators. We demonstrate
that the NOLSM method, which is at the center point of each AIMD step, is able
to achieve a sustained performance of 324 PFLOP/s in mixed FP16/FP32 precision
corresponding to an efficiency of 67.7% when running on 1536 NVIDIA A100 GPUs.
AU - Schade, Robert
AU - Kenter, Tobias
AU - Elgabarty, Hossam
AU - Lass, Michael
AU - Schütt, Ole
AU - Lazzaro, Alfio
AU - Pabst, Hans
AU - Mohr, Stephan
AU - Hutter, Jürg
AU - Kühne, Thomas
AU - Plessl, Christian
ID - 21732
TI - Enabling Electronic Structure-Based Ab-Initio Molecular Dynamics Simulations with Hundreds of Millions of Atoms
ER -
TY - JOUR
AB - CP2K is an open source electronic structure and molecular dynamics software package to perform atomistic simulations of solid-state, liquid, molecular, and biological systems. It is especially aimed at massively parallel and linear-scaling electronic structure methods and state-of-theart ab initio molecular dynamics simulations. Excellent performance for electronic structure calculations is achieved using novel algorithms implemented for modern high-performance computing systems. This review revisits the main capabilities of CP2K to perform efficient and accurate electronic structure simulations. The emphasis is put on density functional theory and multiple post–Hartree–Fock methods using the Gaussian and plane wave approach and its augmented all-electron extension.
AU - Kühne, Thomas
AU - Iannuzzi, Marcella
AU - Ben, Mauro Del
AU - Rybkin, Vladimir V.
AU - Seewald, Patrick
AU - Stein, Frederick
AU - Laino, Teodoro
AU - Khaliullin, Rustam Z.
AU - Schütt, Ole
AU - Schiffmann, Florian
AU - Golze, Dorothea
AU - Wilhelm, Jan
AU - Chulkov, Sergey
AU - Mohammad Hossein Bani-Hashemian, Mohammad Hossein Bani-Hashemian
AU - Weber, Valéry
AU - Borstnik, Urban
AU - Taillefumier, Mathieu
AU - Jakobovits, Alice Shoshana
AU - Lazzaro, Alfio
AU - Pabst, Hans
AU - Müller, Tiziano
AU - Schade, Robert
AU - Guidon, Manuel
AU - Andermatt, Samuel
AU - Holmberg, Nico
AU - Schenter, Gregory K.
AU - Hehn, Anna
AU - Bussy, Augustin
AU - Belleflamme, Fabian
AU - Tabacchi, Gloria
AU - Glöß, Andreas
AU - Lass, Michael
AU - Bethune, Iain
AU - Mundy, Christopher J.
AU - Plessl, Christian
AU - Watkins, Matt
AU - VandeVondele, Joost
AU - Krack, Matthias
AU - Hutter, Jürg
ID - 16277
IS - 19
JF - The Journal of Chemical Physics
TI - CP2K: An electronic structure and molecular dynamics software package - Quickstep: Efficient and accurate electronic structure calculations
VL - 152
ER -
TY - CONF
AB - Electronic structure calculations based on density-functional theory (DFT)
represent a significant part of today's HPC workloads and pose high demands on
high-performance computing resources. To perform these quantum-mechanical DFT
calculations on complex large-scale systems, so-called linear scaling methods
instead of conventional cubic scaling methods are required. In this work, we
take up the idea of the submatrix method and apply it to the DFT computations
in the software package CP2K. For that purpose, we transform the underlying
numeric operations on distributed, large, sparse matrices into computations on
local, much smaller and nearly dense matrices. This allows us to exploit the
full floating-point performance of modern CPUs and to make use of dedicated
accelerator hardware, where performance has been limited by memory bandwidth
before. We demonstrate both functionality and performance of our implementation
and show how it can be accelerated with GPUs and FPGAs.
AU - Lass, Michael
AU - Schade, Robert
AU - Kühne, Thomas
AU - Plessl, Christian
ID - 16898
T2 - Proc. International Conference for High Performance Computing, Networking, Storage and Analysis (SC)
TI - A Submatrix-Based Method for Approximate Matrix Function Evaluation in the Quantum Chemistry Code CP2K
ER -
TY - JOUR
AB - In scientific computing, the acceleration of atomistic computer simulations by means of custom hardware is finding ever-growing application. A major limitation, however, is that the high efficiency in terms of performance and low power consumption entails the massive usage of low precision computing units. Here, based on the approximate computing paradigm, we present an algorithmic method to compensate for numerical inaccuracies due to low accuracy arithmetic operations rigorously, yet still obtaining exact expectation values using a properly modified Langevin-type equation.
AU - Rengaraj, Varadarajan
AU - Lass, Michael
AU - Plessl, Christian
AU - Kühne, Thomas
ID - 12878
IS - 2
JF - Computation
TI - Accurate Sampling with Noisy Forces from Approximate Computing
VL - 8
ER -
TY - CONF
AU - Groth, Stefan
AU - Grünewald, Daniel
AU - Teich, Jürgen
AU - Hannig, Frank
ID - 16852
T2 - Proceedings of the 17th ACM International Conference on Computing Frontiers (CF '2020)
TI - A Runtime System for Finite Element Methods in a Partitioned Global Address Space
ER -
TY - JOUR
AU - Platzner, Marco
AU - Plessl, Christian
ID - 12871
JF - Informatik Spektrum
SN - 0170-6012
TI - FPGAs im Rechenzentrum
ER -
TY - JOUR
AU - Riebler, Heinrich
AU - Vaz, Gavin Francis
AU - Kenter, Tobias
AU - Plessl, Christian
ID - 7689
IS - 2
JF - ACM Trans. Archit. Code Optim. (TACO)
KW - htrop
TI - Transparent Acceleration for Heterogeneous Platforms with Compilation to OpenCL
VL - 16
ER -
TY - CONF
AB - Stratix 10 FPGA cards have a good potential for the acceleration of HPC workloads since the Stratix 10 product line introduces devices with a large number of DSP and memory blocks. The high level synthesis of OpenCL codes can play a fundamental role for FPGAs in HPC, because it allows to implement different designs with lower development effort compared to hand optimized HDL. However, Stratix 10 cards are still hard to fully exploit using the Intel FPGA SDK for OpenCL. The implementation of designs with thousands of concurrent arithmetic operations often suffers from place and route problems that limit the maximum frequency or entirely prevent a successful synthesis. In order to overcome these issues for the implementation of the matrix multiplication, we formulate Cannon's matrix multiplication algorithm with regard to its efficient synthesis within the FPGA logic. We obtain a two-level block algorithm, where the lower level sub-matrices are multiplied using our Cannon's algorithm implementation. Following this design approach with multiple compute units, we are able to get maximum frequencies close to and above 300 MHz with high utilization of DSP and memory blocks. This allows for performance results above 1 TeraFLOPS.
AU - Gorlani, Paolo
AU - Kenter, Tobias
AU - Plessl, Christian
ID - 15478
T2 - Proceedings of the International Conference on Field-Programmable Technology (FPT)
TI - OpenCL Implementation of Cannon's Matrix Multiplication Algorithm on Intel Stratix 10 FPGAs
ER -
TY - JOUR
AB - We address the general mathematical problem of computing the inverse p-th
root of a given matrix in an efficient way. A new method to construct iteration
functions that allow calculating arbitrary p-th roots and their inverses of
symmetric positive definite matrices is presented. We show that the order of
convergence is at least quadratic and that adaptively adjusting a parameter q
always leads to an even faster convergence. In this way, a better performance
than with previously known iteration schemes is achieved. The efficiency of the
iterative functions is demonstrated for various matrices with different
densities, condition numbers and spectral radii.
AU - Richters, Dorothee
AU - Lass, Michael
AU - Walther, Andrea
AU - Plessl, Christian
AU - Kühne, Thomas
ID - 21
IS - 2
JF - Communications in Computational Physics
TI - A General Algorithm to Calculate the Inverse Principal p-th Root of Symmetric Positive Definite Matrices
VL - 25
ER -
TY - THES
AU - Vaz, Gavin Francis
ID - 14849
TI - Using Just-in-Time Code Generation to Transparently Accelerate Applications in Heterogeneous Systems
ER -
TY - JOUR
AU - Mertens, Jan Cedric
AU - Boschmann, Alexander
AU - Schmidt, M.
AU - Plessl, Christian
ID - 6516
IS - 4
JF - Sports Engineering
SN - 1369-7072
TI - Sprint diagnostic with GPS and inertial sensor fusion
VL - 21
ER -
TY - GEN
AB - Molecular Dynamic (MD) simulations are computationally intensive and accelerating them using specialized hardware is a topic of investigation in many studies. One of the routines in the critical path of MD simulations is the three-dimensional Fast Fourier Transformation (FFT3d). The potential in accelerating FFT3d using hardware is usually bound by bandwidth and memory. Therefore, designing a high throughput solution for an FPGA that overcomes this problem is challenging.
In this thesis, the feasibility of offloading FFT3d computations to FPGA implemented using OpenCL is investigated. In order to mask the latency in memory access, an FFT3d that overlaps computation with communication is designed. The implementa- tion of this design is synthesized for the Arria 10 GX 1150 FPGA and evaluated with the FFTW benchmark. Analysis shows a better performance using FPGA over CPU for larger FFT sizes, with the 643 FFT showing a 70% improvement in runtime using FPGAs.
This FFT3d design is integrated with CP2K to explore the potential in accelerating molecular dynamic simulations. Evaluation of CP2K simulations using FPGA shows a 41% improvement in runtime in FFT3d computations over CPU for larger FFT3d designs.
AU - Ramaswami, Arjun
ID - 5417
KW - FFT: FPGA
KW - CP2K
KW - OpenCL
TI - Accelerating Molecular Dynamic Simulations by Offloading Fast Fourier Transformations to FPGA
ER -
TY - JOUR
AB - Approximate computing has shown to provide new ways to improve performance
and power consumption of error-resilient applications. While many of these
applications can be found in image processing, data classification or machine
learning, we demonstrate its suitability to a problem from scientific
computing. Utilizing the self-correcting behavior of iterative algorithms, we
show that approximate computing can be applied to the calculation of inverse
matrix p-th roots which are required in many applications in scientific
computing. Results show great opportunities to reduce the computational effort
and bandwidth required for the execution of the discussed algorithm, especially
when targeting special accelerator hardware.
AU - Lass, Michael
AU - Kühne, Thomas
AU - Plessl, Christian
ID - 20
IS - 2
JF - Embedded Systems Letters
SN - 1943-0663
TI - Using Approximate Computing for the Calculation of Inverse Matrix p-th Roots
VL - 10
ER -
TY - GEN
AU - Filmwala, Tasneem
ID - 5414
TI - Study Effects of Approximation on Conjugate Gradient Algorithm and Accelerate it on FPGA Platform
ER -
TY - GEN
AU - Gadewar, Onkar
ID - 5421
TI - Programmable Programs? - Designing FPGA Overlay Architectures with OpenCL
ER -
TY - CONF
AU - Riebler, Heinrich
AU - Vaz, Gavin Francis
AU - Kenter, Tobias
AU - Plessl, Christian
ID - 1204
KW - htrop
SN - 9781450349826
T2 - Proc. ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming (PPoPP)
TI - Automated Code Acceleration Targeting Heterogeneous OpenCL Devices
ER -
TY - CONF
AB - The exploration of FPGAs as accelerators for scientific simulations has so far mostly been focused on small kernels of methods working on regular data structures, for example in the form of stencil computations for finite difference methods. In computational sciences, often more advanced methods are employed that promise better stability, convergence, locality and scaling. Unstructured meshes are shown to be more effective and more accurate, compared to regular grids, in representing computation domains of various shapes. Using unstructured meshes, the discontinuous Galerkin method preserves the ability to perform explicit local update operations for simulations in the time domain. In this work, we investigate FPGAs as target platform for an implementation of the nodal discontinuous Galerkin method to find time-domain solutions of Maxwell's equations in an unstructured mesh. When maximizing data reuse and fitting constant coefficients into suitably partitioned on-chip memory, high computational intensity allows us to implement and feed wide data paths with hundreds of floating point operators. By decoupling off-chip memory accesses from the computations, high memory bandwidth can be sustained, even for the irregular access pattern required by parts of the application. Using the Intel/Altera OpenCL SDK for FPGAs, we present different implementation variants for different polynomial orders of the method. In different phases of the algorithm, either computational or bandwidth limits of the Arria 10 platform are almost reached, thus outperforming a highly multithreaded CPU implementation by around 2x.
AU - Kenter, Tobias
AU - Mahale, Gopinath
AU - Alhaddad, Samer
AU - Grynko, Yevgen
AU - Schmitt, Christian
AU - Afzal, Ayesha
AU - Hannig, Frank
AU - Förstner, Jens
AU - Plessl, Christian
ID - 1588
KW - tet_topic_hpc
T2 - Proc. Int. Symp. on Field-Programmable Custom Computing Machines (FCCM)
TI - OpenCL-based FPGA Design to Accelerate the Nodal Discontinuous Galerkin Method for Unstructured Meshes
ER -
TY - CONF
AB - We present the submatrix method, a highly parallelizable method for the approximate calculation of inverse p-th roots of large sparse symmetric matrices which are required in different scientific applications. Following the idea of Approximate Computing, we allow imprecision in the final result in order to utilize the sparsity of the input matrix and to allow massively parallel execution. For an n x n matrix, the proposed algorithm allows to distribute the calculations over n nodes with only little communication overhead. The result matrix exhibits the same sparsity pattern as the input matrix, allowing for efficient reuse of allocated data structures.
We evaluate the algorithm with respect to the error that it introduces into calculated results, as well as its performance and scalability. We demonstrate that the error is relatively limited for well-conditioned matrices and that results are still valuable for error-resilient applications like preconditioning even for ill-conditioned matrices. We discuss the execution time and scaling of the algorithm on a theoretical level and present a distributed implementation of the algorithm using MPI and OpenMP. We demonstrate the scalability of this implementation by running it on a high-performance compute cluster comprised of 1024 CPU cores, showing a speedup of 665x compared to single-threaded execution.
AU - Lass, Michael
AU - Mohr, Stephan
AU - Wiebeler, Hendrik
AU - Kühne, Thomas
AU - Plessl, Christian
ID - 1590
KW - approximate computing
KW - linear algebra
KW - matrix inversion
KW - matrix p-th roots
KW - numeric algorithm
KW - parallel computing
SN - 978-1-4503-5891-0/18/07
T2 - Proc. Platform for Advanced Scientific Computing (PASC) Conference
TI - A Massively Parallel Algorithm for the Approximate Calculation of Inverse p-th Roots of Large Sparse Matrices
ER -
TY - JOUR
AU - Schumacher, Jörn
AU - Plessl, Christian
AU - Vandelli, Wainer
ID - 1589
JF - Journal of Physics: Conference Series
TI - High-Throughput and Low-Latency Network Communication with NetIO
VL - 898
ER -
TY - CONF
AB - Compared to classical HDL designs, generating FPGA with high-level synthesis from an OpenCL specification promises easier exploration of different design alternatives and, through ready-to-use infrastructure and common abstractions for host and memory interfaces, easier portability between different FPGA families. In this work, we evaluate the extent of this promise. To this end, we present a parameterized FDTD implementation for photonic microcavity simulations. Our design can trade-off different forms of parallelism and works for two independent OpenCL-based FPGA design flows. Hence, we can target FPGAs from different vendors and different FPGA families. We describe how we used pre-processor macros to achieve this flexibility and to work around different shortcomings of the current tools. Choosing the right design configurations, we are able to present two extremely competitive solutions for very different FPGA targets, reaching up to 172 GFLOPS sustained performance. With the portability and flexibility demonstrated, code developers not only avoid vendor lock-in, but can even make best use of real trade-offs between different architectures.
AU - Kenter, Tobias
AU - Förstner, Jens
AU - Plessl, Christian
ID - 1592
KW - tet_topic_hpc
T2 - Proc. Int. Conf. on Field Programmable Logic and Applications (FPL)
TI - Flexible FPGA design for FDTD using OpenCL
ER -