Build instructions:
Dependencies:
- Boost 1.62+
Windows:
Windows builds made by us are available here: https://github.com/nicehash/nheqminer/releases
Download and install:
- CUDA SDK (if not needed remove USE_CUDA_TROMP and USE_CUDA_DJEZO from nheqminer Preprocessor definitions under Properties > C/C++ > Preprocessor)
- Visual Studio 2013 Community: https://www.visualstudio.com/en-us/news/releasenotes/vs2013-community-vs
- Visual Studio Update 5 installed
- 64 bit version only
Open nheqminer.sln under nheqminer/nheqminer.sln and build. You will have to build ReleaseSSE2 cpu_tromp project first, then Release7.5 cuda_tromp project, then select Release and build all.
Enabled solvers:
- USE_CPU_TROMP
- USE_CPU_XENONCAT
- USE_CUDA_TROMP
- USE_CUDA_DJEZO
If you don't wan't to build with all solvlers you can go to nheqminer Properties > C/C++ > Preprocessor > Preprocessor Definitions and remove the solver you don't need.
Linux
Work in progress. Working solvers CPU_TROMP, CPU_XENONCAT, CUDA_TROMP, CUDA_DJEZO
General instructions:
-
Install CUDA SDK v8 (make sure you have cuda libraries in LD_LIBRARY_PATH and cuda toolkit bins in PATH)
- example on Ubuntu:
- LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda-8.0/lib64:/usr/local/cuda-8.0/lib64/stubs"
- PATH="$PATH:/usr/local/cuda-8.0/"
- PATH="$PATH:/usr/local/cuda-8.0/bin"
-
Use Boost 1.62+ (if it is not available from the repos you will have to download and build it yourself)
-
CMake v3.5 (if it is not available from the repos you will have to download and build it yourself)
-
Currently support only static building (CPU_XENONCAT, CUDA_DJEZO are enabled by default, check CMakeLists.txt in nheqminer root folder)
-
If not on Ubuntu make sure you have fasm installed and accessible in PATH
-
After that open the terminal and run the following commands:
git clone https://github.com/nicehash/nheqminer.git
- Generating asm object file:
- On Ubuntu:
cd nheqminer/cpu_xenoncat/asm_linux/
sh assemble.sh
- bundeled fasm not compatible:
- delete/replace (inside nheqminer/cpu_xenoncat/asm_linux/ directory) with fasm binary compatible with your distro
cd nheqminer/cpu_xenoncat/asm_linux/
sh assemble.sh
- On Ubuntu:
cd ../../../
mkdir build && cd build
cmake ../nheqminer
make -j $(nproc)
Run instructions:
Parameters: -h Print this help and quit -l [location] Stratum server:port -u [username] Username (bitcoinaddress) -a [port] Local API port (default: 0 = do not bind) -d [level] Debug print level (0 = print all, 5 = fatal only, default: 2) -b [hashes] Run in benchmark mode (default: 200 iterations)
CPU settings -t [num_thrds] Number of CPU threads -e [ext] Force CPU ext (0 = SSE2, 1 = AVX, 2 = AVX2)
NVIDIA CUDA settings -ci CUDA info -cd [devices] Enable CUDA mining on spec. devices -cb [blocks] Number of blocks -ct [tpb] Number of threads per block Example: -cd 0 2 -cb 12 16 -ct 64 128
If run without parameters, miner will start mining with 75% of available logical CPU cores. Use parameter -h to learn about available parameters:
Example to run benchmark on your CPU:
nheqminer -b
Example to mine on your CPU with your own BTC address and worker1 on NiceHash USA server:
nheqminer -l equihash.usa.nicehash.com:3357 -u YOUR_BTC_ADDRESS_HERE.worker1
Example to mine on your CPU with your own BTC address and worker1 on EU server, using 6 threads:
nheqminer -l equihash.eu.nicehash.com:3357 -u YOUR_BTC_ADDRESS_HERE.worker1 -t 6
Note: if you have a 4-core CPU with hyper threading enabled (total 8 threads) it is best to run with only 6 threads (experimental benchmarks shows that best results are achieved with 75% threads utilized)
Example to mine on your CPU as well on your CUDA GPUs with your own BTC address and worker1 on EU server, using 6 CPU threads and 2 CUDA GPUs:
nheqminer -l equihash.eu.nicehash.com:3357 -u YOUR_BTC_ADDRESS_HERE.worker1 -t 6 -cd 0 1