dvfs_gpu

An autotuning framework using DVFS on GPUs for the mining of cryptocurrencies. A detailed description (in German language) of the framework can be found in the doc folder. The framework consists of a main program profit_opitmization and several other programs representing single components of the main program. The individual programs are:

• profit_optimization: Main program with frequency optimization and calculation/monitoring of energy costs and mining earnings. GPUs, currencies and optimization algorithms that should be used are configurable through JSON files. The available command line options with a short description are listed when executing profit_optimization --help.
• freq_optimization: Optimization of frequencies with specified algorithm for a specified currency on a specified GPU. Offline as well as online optimization possible. The available command line options with a short description are listed when executing freq_optimization --help.
• freq_exhaustive: Testing of all configurable frequencies on a specified GPU for a specified currency. Offline as well as online benchmarks possible. The available command line options with a short description are listed when executing freq_exhaustive --help.
• nvmlOC: Setting of frequencies on a specified GPU using the NVML library. An usage message is displayed when executing nvmlOC.
• nvapiOC: Setting of frequencies on a specified GPU using the NVAPI (Windows) or NV-Control X (Linux) library. An usage message is displayed when executing nvapiOC.

Usage notes

To be able to change GPU frequencies the framework requires root (Linux) or administrator (Windows) privileges. Please note that overclocking frequencies too much can damage your hardware. Usage is at your own risk.
When using the framework under Linux the coolbits option in the X-Configuration (/etc/X11/xorg.conf) has to be set for each GPU to use (see here). Moreover, when the Linux computer is used remotely over ssh the script start_remote_X.sh (folder utils) has to be executed with the source command before program start. This script stops the running X-Server and starts a new one on the display of the remote computer.

Add new currency

To make a new currency available for usage with the framework a corresponding section has to be added to the currency configuration. At program start the modified currency configuration has to be specified using the corresponding command line option. If no currency configuration is specified a default one is created and used.

Installation guide

Framework - Windows

To build the framework under Windows the following steps are necessary:

1. Install Visual Studio 2015 or newer. Alternatively use MinGW 5 or newer. In the case of MinGW add the path to the binaries (gcc.exe, g++.exe, etc.) to the system environment variable PATH and rename mingw32-make.exe in make.exe. With MinGW a suitable IDE is CLion or the Qt Creator, because of the built-in CMake-Support.
2. Install Cuda Toolkit 9.0 or newer as well as the latest NVIDIA driver. Add the path to the binaries (nvcc.exe, nvidia-smi.exe, etc.) to the system environment variable PATH.
3. Install CMake 3.2 or newer. Add the path to the binaries (cmake.exe) to the system environment variable PATH.
4. Install Git for Windows. During the selection of the shell in the installer choose the Windows-Shell, not the preconfigured MinTTY-Shell. Add the path to the binaries (git.exe, bash.exe and sh.exe) to the system environment variable PATH.
5. Checkout the repository at the desired location with git clone https://github.com/UBT-AI2/dvfs_gpu.git.
6. In the directory of the Top-Level-CMakeLists.txt execute the following:
   • mkdir build && cd build
   • cmake -G "MinGW Makefiles" .. for MinGW or cmake .. for Visual Studio.
   • make for MinGW or open generated Visual Studio Solution and build project.
7. Alternatively directly open the Top-Level-CMakeLists.txt in an IDE with CMake-Support (Visual Studio 2017, CLion, Qt Creator) and build the project.

Framework - Linux

To build the framework under Linux the following steps are necessary:

1. Install Cuda Toolkit 9.0 or newer as well as the latest NVIDIA driver. Make sure the g++ compiler version 5 or newer is installed.
2. Install dependencies with the script intstall_linux_dependencies.sh (folder utils).
3. Checkout the repository at the desired location with git clone https://github.com/UBT-AI2/dvfs_gpu.git.
4. In the directory of the Top-Level-CMakeLists.txt execute the following:
   • mkdir build && cd build
   • cmake ..
   • make

Miner

For the default used currencies of the framework there are prebuilt binaries of mining software available for Windows and Linux. To rebuild the miners with available source code execute the script init-submodules.sh (folder utils/submodule-stuff). This script initializes the Git-Submodules in the miner directory and applies the available patches to them. For a guide to build the individual miners see the corresponding repositorities of the miners.