Power Aware Computing on GPUs

Kasichayanula, Kiran; Terpstra, Dan; Łuszczek, Piotr; Tomov, Stanimire; Moore, Shirley; Peterson, Gregory D.

doi:10.1109/saahpc.2012.26

Cited by 63 publications

(34 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The memory power consumes around 25% of total GPU power in various reports [5]. Studies have examined the power consumption of different components of GPU [18,19]. The power consumption of device memory is much higher than on-chip memory.…”

Section: Profiling Gpu Power With Nvmlmentioning

confidence: 99%

A Step towards Energy Efficient Computing: Redesigning a Hydrodynamic Application on CPU-GPU

Dong

Dobrev

Kolev

et al. 2014

2014 IEEE 28th International Parallel and Distributed Processing Symposium

View full text Add to dashboard Cite

Power and energy consumption are becoming an increasing concern in high performance computing. Compared to multi-core CPUs, GPUs have a much better performance per watt. In this paper we discuss efforts to redesign the most computation intensive parts of BLAST, an application that solves the equations for compressible hydrodynamics with high order finite elements, using GPUs [10,1]. In order to exploit the hardware parallelism of GPUs and achieve high performance, we implemented custom linear algebra kernels. We intensively optimized our CUDA kernels by exploiting the memory hierarchy, which exceed the vendor's library routines substantially in performance. We proposed an autotuning technique to adapt our CUDA kernels to the orders of the finite element method. Compared to a previous base implementation, our redesign and optimization lowered the energy consumption of the GPU in two aspects: 60% less time to solution and 10% less power required. Compared to the CPU-only solution, our GPU accelerated BLAST obtained a 2.5× overall speedup and 1.42× energy efficiency (greenup) using 4th order (Q4) finite elements, and a 1.9× speedup and 1.27× greenup using 2nd order (Q2) finite elements.

show abstract

Section: Profiling Gpu Power With Nvmlmentioning

confidence: 99%

A Step towards Energy Efficient Computing: Redesigning a Hydrodynamic Application on CPU-GPU

Dong

Dobrev

Kolev

et al. 2014

2014 IEEE 28th International Parallel and Distributed Processing Symposium

View full text Add to dashboard Cite

show abstract

“…Nvidia offers utilities to measure power accurately using the Nvidia Management Library (NVML) whereby power is estimated with milliwatt precision within a range of ±5 W [22]. The tool nvidia-smi gives information about power consumption, temperature and memory occupation but the sample rate is rather low.…”

Section: A Methodology Of Power Measurementmentioning

confidence: 99%

Energy-Efficient Collective Reduce and Allreduce Operations on Distributed GPUs

Oden

Klenk

Fröning

2014

2014 14th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing

View full text Add to dashboard Cite

GPUs gain high popularity in High Performance Computing, due to their massive parallelism and high performance per Watt. Despite their popularity, data transfer between multiple GPUs in a cluster remains a problem. Most communication models require the CPU to control the data flow; also intermediate staging copies to host memory are often inevitable. These two facts lead to higher CPU and memory utilization. As a result, overall performance decreases and power consumption increases.Collective operations like reduce and allreduce are very common in scientific simulations and also very sensitive to performance. Due to their massive parallelism, GPUs are very suitable for such operations, but they only excel in performance if they can process the problem in-core. Global GPU Address Spaces (GGAS) enable a direct GPU-to-GPU communication for heterogeneous clusters, which is completely in-line with the GPUs thread-collective execution model and does not require CPU assistance or staging copies in host memory. As we will see, GGAS helps to process collective operations among distributed GPUs in-core.In this paper, we introduce the implementation and optimization of collective reduce and allreduce operations using GGAS as a communication model. Compared to message passing, we get a speedup of 1.7x for small data sizes. A detailed analysis based on power measurements of CPU, host memory and GPU reveals that GGAS as communication model not only saves cycles, also the power and energy consumption is reduced dramatically. For instance, for an allreduce operation half of the energy can be saved by the reduced the power consumption in combination with the lower run time.

show abstract

“…Recent NVIDIA GPUs can report power usage via the NVIDIA Management Library (NVML). Indeed, the authors of [23] use NVML to monitor the power and energy consumption of applications. The RAPL energy counters are able to measure the instantaneous power of these Intel processors at a sampling rate equal to 1,000 S/s and with a fairly high accuracy (less than 1 W).…”

Section: Related Workmentioning

confidence: 99%

Assessing Power Monitoring Approaches for Energy and Power Analysis of Computers

Diouri

Dolz

Glück

et al. 2014

Sustainable Computing: Informatics and Systems

View full text Add to dashboard Cite

Large-scale distributed systems (e.g., datacenters, HPC systems, clouds, large-scale networks, etc.) consume and will consume enormous amounts of energy. Therefore, accurately monitoring the power dissipation and energy consumption of these systems is more unavoidable. The main novelty of this contribution is the analysis and evaluation of different external and internal power monitoring devices tested using two different computing systems, a server and a desktop machine. Furthermore, we provide experimental results for a variety of benchmarks which intensively exercise the main components (CPU, Memory, HDDs, and NICs) of the target platforms to validate the accuracy of the equipment in terms of power dissipation and energy consumption. On the other hand, we also evaluate three different power measurement interfaces available on current architecture generations. Thanks to the high sampling rate and to the different measured lines, the internal wattmeters allow an improved visualization of some power fluctuations. However, a high sampling rate is not always necessary to understand the evolution of the power consumption during the execution of a benchmark.

show abstract

Power Aware Computing on GPUs

Cited by 63 publications

References 9 publications

A Step towards Energy Efficient Computing: Redesigning a Hydrodynamic Application on CPU-GPU

A Step towards Energy Efficient Computing: Redesigning a Hydrodynamic Application on CPU-GPU

Energy-Efficient Collective Reduce and Allreduce Operations on Distributed GPUs

Assessing Power Monitoring Approaches for Energy and Power Analysis of Computers

Contact Info

Product

Resources

About