DVFS and Timing Optimization on GPU for Data Center Computation
DOI:
https://doi.org/10.21831/jraee.v2i1.556Keywords:
Data Center, GPU, Computation, DVFS, Timing OptimizationAbstract
Data center computing requires efficient GPU support, both in terms of functionality and power consumption. GPU performance efficiency can be reduced due to high power usage and reduced GPU work stability. So it requires an analysis of computational performance and power efficiency to improve performance and reduce power usage. Core voltage, core frequency, and memory timings are parameters that affect the efficiency of computing performance, power efficiency, and stability. Increasing computational efficiency and GPU power with the effect of modifying parameters can be done through the Basic Input-Output System (BIOS). This study analyzes the efficiency of computational performance by optimizing memory timings and analyzing power efficiency and stability by modifying the DVFS algorithm. Tests are carried out using computational benchmarks commonly used in data centers including the tessellation algorithm, rendering, image processing, pi calculation, image stitching, deep learning, molecular simulation, and N-body. The efficiency of computing performance and GPU power efficiency can be increased by optimizing memory timings and changing the voltage and frequency values on DVFS. Increased performance efficiency ranged from 33.3% to 66.7% and power efficiency increased from 19.9% to 32.6%. Modification of the DVFS voltage state can increase voltage stability and GPU core frequency stability.
Downloads
References
Y. Arafa, A. -H. A. Badawy, G. Chennupati, N. Santhi and S. Eidenbenz, "PPT-GPU: Scalable GPU Performance Modeling," in IEEE Computer Architecture Letters, vol. 18, no. 1, pp. 55-58, 1 Jan.-June 2019, doi: 10.1109/LCA.2019.2904497
S. Najam, J. Ahmed, S. Masood and C. M. Ahmed, "Run-Time Resource Management Controller for Power Efficiency of GP-GPU Architecture," in IEEE Access, vol. 7, pp. 25493-25505, 2019, doi: 10.1109/ACCESS.2019.2901010.
Y. Ma, J. Zhou, T. Chantem, R. P. Dick, S. Wang and X. S. Hu, "Improving Reliability of Soft Real-Time Embedded Systems on Integrated CPU and GPU Platforms," in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 39, no. 10, pp. 2218-2229, Oct. 2020, doi: 10.1109/TCAD.2019.2940681.
M. Smith, L. Zhao, J. Cordova, X. Jiang and M. Ebrahimi, "Energy-Efficient GPU-Intensive Workload Scheduling for Data Centers," 2023 International Conference on Machine Learning and Applications (ICMLA), Jacksonville, FL, USA, 2023, pp. 1735-1740, doi: 10.1109/ICMLA58977.2023.00263.
S. . -K. Shekofteh, H. Noori, M. Naghibzadeh, H. Fröning and H. S. Yazdi, "cCUDA: Effective Co- Scheduling of Concurrent Kernels on GPUs," in IEEE Transactions on Parallel and Distributed Systems, vol. 31, no. 4, pp. 766-778, 1 April 2020, doi: 10.1109/TPDS.2019.2944602.
C. A. García-Rodríguez, P. Quinto-Diez, J. A. Jiménez-Bernal, L. A. R. -D. León and A. Reyes-León, "Waste Heat Recovery System Applied to a High-Performance Video Card," in IEEE Access, vol. 8, pp. 6272-6281, 2020, doi: 10.1109/ACCESS.2020.2964207.
J. Guerreiro, A. Ilic, N. Roma and P. Tomás, "Modeling and Decoupling the GPU Power Consumption for Cross-Domain DVFS," in IEEE Transactions on Parallel and Distributed Systems, vol. 30, no. 11, pp. 2494-2506, 1 Nov. 2019, doi: 10.1109/TPDS.2019.2917181.
S. M. Nabavinejad, S. Reda and M. Ebrahimi, "Coordinated Batching and DVFS for DNN Inference on GPU Accelerators," in IEEE Transactions on Parallel and Distributed Systems, vol. 33, no. 10, pp. 2496- 2508, 1 Oct. 2022, doi: 10.1109/TPDS.2022.3144614.
H. Liu, S. Liu, C. Wen and W. E. Wong, "TBEM: Testing-Based GPU-Memory Consumption Estimation for Deep Learning," in IEEE Access, vol. 10, pp. 39674-39680, 2022, doi: 10.1109/ACCESS.2022.3164510.
C. Zhang, F. Zhang, X. Guo, B. He, X. Zhang and X. Du, "iMLBench: A Machine Learning Benchmark Suite for CPU-GPU Integrated Architectures," in IEEE Transactions on Parallel and Distributed Systems, vol. 32, no. 7, pp. 1740-1752, 1 July 2021, doi: 10.1109/TPDS.2020.3046870.
