Quantifying the impact of GPUs on performance and energy efficiency in HPC clusters

Enos, Jeremy; Steffen, Craig; Fullop, Joshi; Showerman, Michael; Shi, Guosheng; Esler, K.; Kindratenko, Volodymyr; Stone, John E.; Phillips, J. C.

doi:10.1109/greencomp.2010.5598297

Cited by 37 publications

(30 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…GPUs have become an effective tool for acceleration of a wide variety of computationally demanding molecular modeling applications [4]- [9], and these successes have contributed to the use of GPUs in the top performing supercomputers in the world. The Blue Waters petascale supercomputer at U. Illinois 1 is one of the most recent to incorporate GPUs to improve peak performance, 1 URL: http://www.ncsa.illinois.edu/BlueWaters/ as well as price/performance and performance/watt ratios [10]. Blue Waters is an ideal target platform for large scale molecular visualization and analysis tasks because it is composed of both Cray XE6 (CPU-only) and heterogeneous Cray XK7 (GPU-accelerated) compute nodes, allowing users to run jobs on nodes with the most appropriate mix of processors and memory capacity for the task at hand.…”

Section: Introductionmentioning

confidence: 99%

Early experiences scaling VMD molecular visualization and analysis jobs on blue waters

Stone

Isralewitz

Schulten

2013

2013 Extreme Scaling Workshop (Xsw 2013)

Self Cite

View full text Add to dashboard Cite

Abstract-Petascale molecular dynamics simulations provide a powerful tool for probing the dynamics of cellular processes at atomic and nanosecond resolution not achievable by experimental methods alone. Extraction of details about the dynamics of biomolecules from terabytes of simulation output requires powerful user-extensible molecular analysis and visualization tools. Petascale simulation trajectories are so large that it is now necessary to perform many analysis and visualization tasks formerly handled by off-site computational facilities in-place on the supercomputer itself. We report ongoing progress on porting, tuning, and scaling up the popular molecular visualization and analysis program VMD on the NSF Blue Waters petascale supercomputer. We describe key achievements including algorithmic and memory efficiency improvements, hand-vectorization of key CPU algorithms, new and improved GPU analysis and visualization algorithms, and parallel I/O performance results. We evaluate the performance of VMD for user-developed analysis scripts with the TIMELINE trajectory analysis tool in VMD. Finally, we describe the unique capabilities provided by the Cray XK7 GPU-accelerated compute partition of Blue Waters.

show abstract

Section: Introductionmentioning

confidence: 99%

Early experiences scaling VMD molecular visualization and analysis jobs on blue waters

Stone

Isralewitz

Schulten

2013

2013 Extreme Scaling Workshop (Xsw 2013)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Heterogeneous hardware platforms, combining central processing units (CPUs) and GPUs, allow significant savings of computing time and power consumption for some applications, e.g. [19,13,27] to name but a few.…”

Section: Introductionmentioning

confidence: 99%

Performance of a projection method for incompressible flows on heterogeneous hardware

et al. 2015

View full text Add to dashboard Cite

“…First, in addition to increasing the acquisition costs, the use of accelerators increases maintenance, administration, and space costs [6]. Second, energy consumption is increased, as GPUs are known to be power-hungry devices [7]. Third, GPUs in such a cluster may present a relatively low utilization rate, given that it is unlikely that all the accelerators in the cluster will be used all the time, since few applications feature such an extreme dataconcurrency degree.…”

Section: Introductionmentioning

confidence: 99%

Influence of InfiniBand FDR on the performance of remote GPU virtualization

Reaño

Mayo

Quintana‐Ortí

et al. 2013

2013 IEEE International Conference on Cluster Computing (CLUSTER)

View full text Add to dashboard Cite

Abstract-The use of GPUs to accelerate general-purpose scientific and engineering applications is mainstream today, but their adoption in current high-performance computing clusters is impaired primarily by acquisition costs and power consumption. Therefore, the benefits of sharing a reduced number of GPUs among all the nodes of a cluster can be remarkable for many applications. This approach, usually referred to as remote GPU virtualization, aims at reducing the number of GPUs present in a cluster, while increasing their utilization rate.The performance of the interconnection network is key to achieving reasonable performance results by means of remote GPU virtualization. To this end, several networking technologies with throughput comparable to that of PCI Express have appeared recently. In this paper we analyze the influence of InfiniBand FDR on the performance of remote GPU virtualization, comparing its impact on a variety of GPU-accelerated applications with other networking technologies, such as InfiniBand QDR and Gigabit Ethernet. Given the severe limitations of freely available remote GPU virtualization solutions, the rCUDA framework is used as the case study for this analysis. Results show that the new FDR interconnect, featuring higher bandwidth than its predecessors, allows the reduction of the overhead of using GPUs remotely, thus making this approach even more appealing.

show abstract

Quantifying the impact of GPUs on performance and energy efficiency in HPC clusters

Cited by 37 publications

References 10 publications

Early experiences scaling VMD molecular visualization and analysis jobs on blue waters

Early experiences scaling VMD molecular visualization and analysis jobs on blue waters

Performance of a projection method for incompressible flows on heterogeneous hardware

Influence of InfiniBand FDR on the performance of remote GPU virtualization

Contact Info

Product

Resources

About