Non-Conventional AMD GPU Characterization Tool and Methodology

Set of benchmarks ana analysis tools to characterize the response of AMD GPU architecture to different workloads whily using non-conventional (extreme undervoltage) DVFS settings

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes. See deployment for notes on how to deploy the project on a live system.

Prerequisites (for CentOS 7.0)

• Python 3.8
• Anaconda/Miniconda enviroment
• ROCm 3.0
• ROCblas
• MIopen-hip
• MIopen-gemm

Installing

• Create conda enviroment for analysis tool

conda env create -f environment/environment.yml

• ROCm 3.0

sudo yum install rocm-dkms rock-dkms
sudo yum install rocblas
sudo yum install miopen-hip
sudo yum install miopengemm

Running the tests

Use the exploreBenchmark Python script to run a Benchmark on different GPU DVFS configurations

General example for running benchmark

python exploreBenchmark.py -b <location of executable file> -e -c -v -t 10 -lc 7 --step 25 --reset --config <configuration file>

Example for running SFU benchmark

python exploreBenchmark.py -b Benchmarks/Core/ALU/SFU/Float/sp/eventPerformance_SFU_Float_sp -e -c -v -t 10 -lc 7 --step 25 --reset --config configExploreFiles/11.txt

Runing analysis tool

python analysisBenchmark.py -p <location of results files> -c <config file>

Built With

• Python - For running and analysis script
• C++ - Benchmarks
• HIP - AMD Kernels

Authors

• Francisco Mendes - TheEmbbededCoder
• IST - INESC-ID - INESC-ID

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

• Hat tip to anyone whose code was used
• Inspiration
• The professors Nuno Roma and Pedro Tomás for all the discussion and debate over computer architecture