Runs ethminer with all available Nvidia and AMD GPUs. This was tested on a system using Ubuntu 16.04LTS, containing the following specs:

• ASUS B250 Mining Expert
• 8 x GTX 1070 / 8GB GPUs
• 4 x AMD RX580 / 8GB GPUs.
• 32GB RAM
• 1TB Samsung SSD (for fast read/write)

Note: AMD GPUs were recorded as opencl-platform with value 0. It could be different for your system. So check by running ethminer --list-devices and look for the platform-id in the output for AMD (Ellesmere), usually it is 0.

• For Nvidia cards, tunes them by undervolting GPU and increasing memory frequency, increasing mining to 30MH/s+ for Eth
• For AMD cards, the BIOS need to be flashed using Polaris Bios Editor in Windows, on the same machine as the miner (use a stick).

• Create a .account file that contains the Ethereum account alphanumeric string and nickname, like this: <account-address>.<nickname> and put it in the root folder of this project.
• nvidia-smi (comes with Nvidia drivers)
• nvidia-xconfig (comes with Nvidia drivers)
• Install ethminer. Make sure ethminer --list-devices recognizes all your GPUs and lists them. Pay attention to the platform-IDs for AMD cards as that's required in the amd/run_amd.sh script.

To just start the miners, without any monitoring:

$ ./start.sh

Monitoring requires a Python virtual-environment. So, to get started, install pipenv, which wraps pip and virtualenv and setup all the required python dependencies (check Pipfile for an overview) NOTE: do not directly edit Pipfile.lock :

$ pip install pipenv
$ pipenv install

Now start the miners with monitoring:

$ ./start.sh --metrics

• Clone the repo.
• Edit init.sh and add the location for Ethminer in ETHMINER_PATH, eg: /home/foobar/.ethminer/bin
• Set the tune parameters for Nvidia tuning in nvidia/tune_nvidia.sh. The default values should fetch 30MH/s on Nvidia 1070 GTX 8GB, but feel free to tweak that even further with higher MEM_OVERCLOCK values and higher POWER values.
• No AMD tuning is provided here, you have to use Polaris Bios Editor to flash the ROM and get your optimal hashrate. A list of available pre-tested ROMS for MSI RX580 8GB cards are provided in rom/ for use, but use it at your own risk as not every GPU is the same.
• Edit amd/run_amd.sh and set the platform-ID for AMD cards fetched using ethminer --list-devices
• Start mining by runing sudo start.sh, which will first enable and tune your Nvidia cards and then will run ethminer on all available Nvidia and AMD GPUs.
• Fluentd daemon will run as a docker container collecting nvidia and amd hash-rates and reporting that to Influxdb/Grafana.

The start.sh, stop.sh and switchoff_until.sh scripts can be used with crontab as shown below:

# m h  dom mon dow   command

# Start miners and watchdog
@reboot sudo /path/to/eth-runner/start.sh > /path/to/eth-runner/logs/start.log 2>&1

# Shutdown at 7am everyday
00 07 * * * sudo /path/to/eth-runner/stop.sh
02 07 * * * sudo /path/to/eth-runner/shutdown.sh

A prepackaged Grafana + InfluxDB + StatsD Docker container is used help store and display eth-runner metrics. The start* scripts in the metrics/ directory collect the metrics via the underlying drivers such as nvidia-smi and lm-sensors (PySensors). For more details on Metrics, check the Readme here.

A Snapshot of the sample Grafana dashboard looks like below:

NOTE: For default data-polling period (5 seconds), not recommended for leaving on long-term, like overnight. It can be taxing on the GPUs, but helpful for debugging/monitoring. If required for long-term, better to use higher ping-intervals of 1-2 minutes, or even 1 hour.