Počet záznamů: 1
GPU-acceleration of the ELPA2 distributed eigensolver for dense symmetric and hermitian eigenproblems
- 1.
SYSNO ASEP 0539376 Druh ASEP J - Článek v odborném periodiku Zařazení RIV J - Článek v odborném periodiku Poddruh J Článek ve WOS Název GPU-acceleration of the ELPA2 distributed eigensolver for dense symmetric and hermitian eigenproblems Tvůrce(i) Yu, V. W. z. (US)
Moussa, J. (US)
Kůs, Pavel (MU-W) RID, SAI, ORCID
Marek, A. (DE)
Messmer, P. (CH)
Yoon, M. (US)
Lederer, H. (DE)
Blum, V. (US)Číslo článku 107808 Zdroj.dok. Computer Physics Communications. - : Elsevier - ISSN 0010-4655
Roč. 262, May (2021)Poč.str. 12 s. Jazyk dok. eng - angličtina Země vyd. NL - Nizozemsko Klíč. slova CUDA ; dense linear algebra ; eigensolver ; high-performance computing ; parallel computing Vědní obor RIV BA - Obecná matematika Obor OECD Pure mathematics Způsob publikování Omezený přístup Institucionální podpora MU-W - RVO:67985840 UT WOS 000633365000004 EID SCOPUS 85099623870 DOI 10.1016/j.cpc.2020.107808 Anotace The solution of eigenproblems is often a key computational bottleneck that limits the tractable system size of numerical algorithms, among them electronic structure theory in chemistry and in condensed matter physics. Large eigenproblems can easily exceed the capacity of a single compute node, thus must be solved on distributed-memory parallel computers. We here present GPU-oriented optimizations of the ELPA two-stage tridiagonalization eigensolver (ELPA2). On top of cuBLAS-based GPU offloading, we add a CUDA kernel to speed up the back-transformation of eigenvectors, which can be the computationally most expensive part of the two-stage tridiagonalization algorithm. We benchmark the performance of this GPU-accelerated eigensolver on two hybrid CPU–GPU architectures, namely a compute cluster based on Intel Xeon Gold CPUs and NVIDIA Volta GPUs, and the Summit supercomputer based on IBM POWER9 CPUs and NVIDIA Volta GPUs. Consistent with previous benchmarks on CPU-only architectures, the GPU-accelerated two-stage solver exhibits a parallel performance superior to the one-stage counterpart. Finally, we demonstrate the performance of the GPU-accelerated eigensolver developed in this work for routine semi-local KS-DFT calculations comprising thousands of atoms. Pracoviště Matematický ústav Kontakt Jarmila Štruncová, struncova@math.cas.cz, library@math.cas.cz, Tel.: 222 090 757 Rok sběru 2022 Elektronická adresa https://doi.org/10.1016/j.cpc.2020.107808
Počet záznamů: 1