Blue Space: a Dark Forest miner

This is a a miner/explorer for the Dark Forest game. It includes a CPU miner implemented using GMP, and a GPU miner implemented in CUDA.

Installation

We provide binary packages for Ubuntu 20.04, you can find them in the Releases section.

The CUDA version requires you to have CUDA and the Nvidia drivers installed.

Usage

Blue Space can be used in two modes:

--benchmark: run a simple benchmark and exit. This is used for tuning the CUDA parameters,
--stateless: run the explorer web server waiting requests from the game.

To toggle between CPU or GPU mode, use the --cpu or --cuda flags. Note that the --cuda flag is available only if you downloaded the CUDA version of Blue Space.

You can get a list of all available flags with blue-space --help.

This miner is compatible with the default "Remote Explorer" plugin included in the game.

Tuning the CUDA miner

The CUDA miner is very sensitive to the parameters you pass to it. The parameters optimal value is dependant on your card, so it's difficult to provide sensible defaults. When tuning the miner, remember to also change the --size and then change it in the remote explorer plugin as well.

The CUDA related flags are:

--cuda-device: If you have multiple cards, change between them by passing the card index. You can find the card index in Nvidia X Settings. By default this flag is 0.
--cuda-thread-work-size: the number of locations explored by each CUDA thread.
--cuda-block-size: the number of virtual threads in each block. See next option.
--cuda-threads-per-item: the number of threads used for each number. The default value of 8 should always be the optimal one. The effective number of threads in a block is given by: block-size * threads-per-item.

Building

You need the following dependencies to build blue-space. You can install them with your distribution favourite package manager.

CMake
The GNU MP Bignum library, a.k.a libgmp
Boost.Log
Google Test
Google Benchmark
CLI11 1.9. You can download the header to a location on your file system and set CLI11_DIR. If this is not installed, CMake will download it.
Json Rpc Lean. Download the source on your file system, then set the JSONRPCLEAN_INCLUDE_DIR configuration variable to the path to jsonrpc-lean include/ directory. If this is not installed, CMake will download it.

To build the GPU miner, you need the following additional dependencies.

The CUDA Toolkit. Follows for example this step-by-step tutorial.
CGBN. You can download the source to a location on your file system and set CGBN_INCLUDE_DIR, otherwise if this is not installed, CMake will download it for you.

On Ubuntu 21.04 you can install all dependencies with:

sudo apt install -y \
    build-essential \
    cmake \
    pkg-config \
    rapidjson-dev \
    libgmp-dev \
    libboost-container-dev \
    libboost-thread-dev \
    libboost-log-dev \
    libgtest-dev \ # optional
    libbenchmark-dev # optional

After you have all dependencies in place, it's time to compile blue-space.

Start by downloading the source to your file system.

git clone https://github.com/long-rock/blue-space.git

Configure cmake, run the following from the blue-space directory.

cmake \
    -S . \ # the source location
    -B builddir \ # the build directory
    -DCMAKE_BUILD_TYPE=Release \ # make release
    -DBUILD_TEST=OFF \ # build tests, requires gtest and benchmark
    -DCUDA_MINER=ON \ # enable cuda miner
    -DCUDA_HOST_COMPILER=/usr/bin/g++-10 \ # only needed if you have g++ 11 on your system
    -DCMAKE_CUDA_ARCHITECTURES=35 \ # cuda architecture, depends on your gpu

Finally, build blue-space.

cmake --build builddir

The blue-space binary is builddir/blue-space/blue-space.

License

Licensed under the GNU General Public License, Version 3.

spacemandev-git / blue-space-arena