Efficiency Analysis of Intel, AMD and Nvidia 64-Bit Hardware for Memory-Bound Problems: A Case Study of Ab Initio Calculations with VASP

Stegaĭlov, V. V.; Vecher, Vyacheslav S.

doi:10.1007/978-3-319-78054-2_8

Cited by 5 publications

(8 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Building an energy‐aware task management mechanism for the MPDATA algorithms on multicore CPUs was proposed by Rojek et al In our previous studies, we considered energy consumption of minicomputers running classical MD benchmarks . This study expands our preliminary results on the VASP analysis …”

Section: Related Workmentioning

confidence: 77%

“…The data point taken from the IBM Power 7 benchmark follows this trend very well. The rightmost point (that corresponds to the benchmark with 1 core of E5‐2620v4 with 1 active memory channel) is not located at the main trend, because in this combination, presumably, all available L3 cache can not be utilized effectively …”

Section: Resultsmentioning

confidence: 99%

“…The main results of this work are obtained on several supercomputers based on different CPUs (Intel Xeon and AMD Epyc) and different types of interconnect (Mellanox Infiniband FDR and EDR, Intel Omni‐Path and Angara). Previously published results for other hardware are used for comparison too …”

Section: Hardware and Softwarementioning

confidence: 99%

“…We use the same test model in VASP as considered previously for the benchmarks of the IBM 775 system . The model corresponds to a crystal of GaAs with 80 atoms in the supercell (Figure ).…”

Section: Hardware and Softwarementioning

confidence: 99%

“…Previously published results for other hardware are used for comparison too. 32,33 As a special case, we present the results for the Nvidia Jetson TX2 minicomputer based on the Nvidia Tegra ''Parker'' SoC with Armv8-A ISA.…”

Section: Hardware and Softwarementioning

confidence: 99%

See 4 more Smart Citations

VASP hits the memory wall: Processors efficiency comparison

Stegaĭlov

Smirnov

Vecher

2019

Concurrency and Computation

Self Cite

View full text Add to dashboard Cite

Summary First‐principles calculations of electronic structure have been one of the most important classes of supercomputer applications for a long time. In this paper, we consider VASP as a de facto standard tool for density functional theory calculations widely used in materials science, condensed matter research, and other related fields. The choice of hardware for the efficient VASP calculations is not easy because of the large number of processor types available. We use the benchmark metric that is based on the balance of the peak floating point performance and the memory bandwidth. This metric gives us the possibility to compare different types of processors. We consider time‐to‐solution and energy‐to‐solution criteria and compare different Intel, AMD, and ARM 64‐bit CPUs and hybrid CPU‐GPU systems based on Nvidia Tesla P100.

show abstract

Section: Related Workmentioning

confidence: 77%

Section: Resultsmentioning

confidence: 99%

Section: Hardware and Softwarementioning

confidence: 99%

“…We use the same test model in VASP as considered previously for the benchmarks of the IBM 775 system . The model corresponds to a crystal of GaAs with 80 atoms in the supercell (Figure ).…”

Section: Hardware and Softwarementioning

confidence: 99%

Section: Hardware and Softwarementioning

confidence: 99%

See 3 more Smart Citations