# make NVCC=/path/to/nvcc
The path and args to the external mining binary can be provided to
chainweb-miner gpu using --miner-path and --miner-args.
In order to amortize GPU startup latency, the mining code can fork a unix
socket server to listen for requests from chainweb-miner. To install:
sudo ./install.sh
This will copy systemd service files to /opt/kadena, and setup systemctl to
run the mining code as a service at boot. If you want to run multiple
instances, you can tweak the service file as desired (and there are
--num-devices and --starting-device flags for the cuda binary).
chainweb-gpu-miner requires the following flags to connect to client:
--client --unix-domain-ns=$NS
The default unix socket namespace is chainweb-gpu-miner0, but if you want to
set up one-chain-per-miner, you'll have to pass --miner-args="--client --unix-domain-ns=$NS" (or something similar) when you are setting up
chainweb-miner.
There is an acceptance test for chainweb mining code in the cwtool binary
provided in the main chainweb-node repository. Run cwtool test-miner for
instructions.