MF-GPOEO

MF-GPOEO is a model-free online energy optimization framework.

./Src is the source code of our work.

./DEPO is our implementation of DEPO for GPU. The source code of DEPO for GPU was still unavailable online when our paper was submitted. (2023/06/16 13:12 UTC, https://projects.task.gda.pl/akrz/split)

./PerformanceMeasurement is our tool to measure GPU power, GPU energy, GPU SM frequency, GPU memory frequency, execution time, CPU power, CPU energy, etc.

Compile and use our work

Compiling

cd ./Src

Set the CUDA installation path (CUDA_DIR) in Makefile.

make -j8

Get the dynamic library (libGPOEO.so) of our work.

Using

# set environment variables
export CUDA_INJECTION64_PATH=<full_path>/libGPOEO.so # the path to libGPOEO.so
export MFGPOEO_CONFIG_PATH=<full_path>/Src/RTX3080Ti.ini # the path to the configuration file
export MFGPOEO_OUTPUT_PATH=<full_path>/log.txt # the path to the log file
export MFGPOEO_WORK_MODE=PID # the work mode

# run a GPU application (running for a period of time, such as 10 minutes)
sudo <full_path>/GPU_Application.bin

The test_MF-GPOEO.sh is an example.

Compile and use DEPO (our implementation)

Compiling

cd ./DEPO

Set the CUDA installation path (CUDA_DIR) in Makefile.

make -j8

Get the dynamic library (libDEPO.so) of DEPO.

Using

# set environment variables
export CUDA_INJECTION64_PATH=<full_path>/libDEPO.so # the path to libDEPO.so
export DEPO_CONFIG_PATH=<full_path>/DEPO/RTX3080Ti.ini # the path to the configuration file
export DEPO_OUTPUT_PATH=<full_path>/log.txt # the path to the log file
export DEPO_WORK_MODE=DEPO # the work mode

# run a GPU application (running for a period of time, such as 10 minutes)
sudo <full_path>/GPU_Application.bin

The test_DEPO.sh is an example.

Compile and use PerformanceMeasurement

Compiling

cd ./PerformanceMeasurement

Set the CUDA installation path (CUDA_DIR) in Makefile.

make -j8

Get the measurement tool (PerfMeasure.bin).

Using

The test.sh is an example.

sudo test.sh

Citation

If you use MF-GPOEO in your research, please cite us as follows:

Wang F, Hao M, Zhang W, et al. Model-Free GPU Online Energy Optimization[J]. IEEE Transactions on Sustainable Computing, 2023. https://github.com/ruixueqingyang/MF-GPOEO

BibTex:

@article{wang2023model,
  title={Model-Free GPU Online Energy Optimization},
  author={Wang, Farui and Hao, Meng and Zhang, Weizhe and Wang, Zheng},
  journal={IEEE Transactions on Sustainable Computing},
  year={2023},
  publisher={IEEE}
}