Power Modeling for Heterogeneous Processors

Diop, Tahir; Jerger, Natalie Enright; Anderson, Jason

doi:10.1145/2576779.2576790

Cited by 6 publications

(4 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our PE model has a maximum percentage error smaller than 2% across all four metrics, while that of the comparable state of the art is in the range of 2%-7% on a single metric [16,17,18,19,20,21]. Moreover, the efficient setup for data extraction and the heuristic search of models help us to reach higher accuracy with less time spent for acquiring data and training the model.…”

Section: Discussion and Key Takeawaysmentioning

confidence: 90%

See 1 more Smart Citation

MLComp: A Methodology for Machine Learning-based Performance Estimation and Adaptive Selection of Pareto-Optimal Compiler Optimization Sequences

Colucci

Juhász

Mosbeck

et al. 2021

2021 Design, Automation &Amp; Test in Europe Conference &Amp; Exhibition (DATE)

View full text Add to dashboard Cite

Embedded systems have proliferated in various consumer and industrial applications with the evolution of Cyber-Physical Systems and the Internet of Things. These systems are subjected to stringent constraints so that embedded software must be optimized for multiple objectives simultaneously, namely reduced energy consumption, execution time, and code size. Compilers offer optimization phases to improve these metrics. However, proper selection and ordering of them depends on multiple factors and typically requires expert knowledge. State-ofthe-art optimizers facilitate different platforms and applications case by case, and they are limited by optimizing one metric at a time, as well as requiring a time-consuming adaptation for different targets through dynamic profiling.To address these problems, we propose the novel MLComp methodology, in which optimization phases are sequenced by a Reinforcement Learning-based policy. Training of the policy is supported by Machine Learning-based analytical models for quick performance estimation, thereby drastically reducing the time spent for dynamic profiling. In our framework, different Machine Learning models are automatically tested to choose the best-fitting one. The trained Performance Estimator model is leveraged to efficiently devise Reinforcement Learning-based multi-objective policies for creating quasioptimal phase sequences.Compared to state-of-the-art estimation models, our Performance Estimator model achieves lower relative error (< 2%) with up to 50× faster training time over multiple platforms and application domains. Our Phase Selection Policy improves execution time and energy consumption of a given code by up to 12% and 6%, respectively. The Performance Estimator and the Phase Selection Policy can be trained efficiently for any target platform and application domain.

show abstract

Section: Discussion and Key Takeawaysmentioning

confidence: 90%

“…State-of-the-art estimation models can be distinguished as MLbased [16,17,18] and formal [19,20,21] ones. ML-based models tend to use accurate sensors and interfaces to estimate the power, and hence require external modifications.…”

Section: Introductionmentioning

confidence: 99%

MLComp: A Methodology for Machine Learning-based Performance Estimation and Adaptive Selection of Pareto-Optimal Compiler Optimization Sequences

Colucci

Juhász

Mosbeck

et al. 2021

2021 Design, Automation &Amp; Test in Europe Conference &Amp; Exhibition (DATE)

View full text Add to dashboard Cite

show abstract

“…Power modeling methodologies have been also studied [Diop et al 2014; Austin and Wright 2014] to analytically study the evolution of processor power as a function of voltage, frequency, and temperature. Skadron et al [2004] in particular studied the role of temperature on leakage, leading to a more realistic power equation.…”

Section: Related Workmentioning

confidence: 99%

Static and Dynamic Frequency Scaling on Multicore CPUs

Bao

Hong

Chunduri

et al. 2016

ACM Trans. Archit. Code Optim.

View full text Add to dashboard Cite

Dynamic Voltage and Frequency Scaling (DVFS) typically adapts CPU power consumption by modifying a processor's operating frequency (and the associated voltage). Typical DVFS approaches include using default strategies such as running at the lowest or the highest frequency or reacting to the CPU's runtime load to reduce or increase frequency based on the CPU usage. In this article, we argue that a compile-time approach to CPU frequency selection is achievable for affine program regions and can significantly outperform runtimebased approaches. We first propose a lightweight runtime approach that can exploit the properties of the power profile specific to a processor, outperforming classical Linux governors such as powersave or on-demand for computational kernels. We then demonstrate that, for affine kernels in the application, a purely compiletime approach to CPU frequency and core count selection is achievable, providing significant additional benefits over the runtime approach. Our framework relies on a one-time profiling of the target CPU, along with a compile-time categorization of loop-based code segments in the application. These are combined to determine at compile-time the frequency and the number of cores to use to execute each affine region to optimize energy or energy-delay product. Extensive evaluation on 60 benchmarks and 5 multi-core CPUs show that our approach systematically outperforms the powersave Linux governor while also improving overall performance.

show abstract

“…Tahir [6] simulates the total power dissipation of the APU and uses the externally connected power measurement unit to measure the power consumption of the processor. We use the Multi2Sim simulator to simulate AMD's EverGreen APUs and obtain performance counter information, including Request to L2 Cache, L2 Cache Miss and DRAM Access, to estimate the power consumption of the processor using the regression model.…”

Section: Figure 3 Details Of Gpgpu-sim [4]mentioning

confidence: 99%

A Survey of Power Consumption Modeling for GPU Architecture

2016

CPCS

View full text Add to dashboard Cite

GPUs are of increasing interests in the multi-core era due to their high computing power. However, the power consumption caused by the rising performance of GPUs has been a general concern. As a consequence, it is becoming an imperative demand to optimize the GPU power consumption, among which the power consumption estimation is one of the important and useful solutions. In this work, we give a survey of the power modeling for GPU. We first introduce the current development of heterogeneous architectures and then summarize the existing modeling techniques for GPU power consumption. The main two types of power modeling could be classified as simulator-based methods and real machine-based methods.

show abstract

Power Modeling for Heterogeneous Processors

Cited by 6 publications

References 21 publications

MLComp: A Methodology for Machine Learning-based Performance Estimation and Adaptive Selection of Pareto-Optimal Compiler Optimization Sequences

MLComp: A Methodology for Machine Learning-based Performance Estimation and Adaptive Selection of Pareto-Optimal Compiler Optimization Sequences

Static and Dynamic Frequency Scaling on Multicore CPUs

A Survey of Power Consumption Modeling for GPU Architecture

Contact Info

Product

Resources

About