lolMiner 1.21
Note: Windows build will follow within next 24 hrs.
Changes (from 1.20)
- slightly improved the performance of Linux zombie mode on Polaris GPUs on medium tune stages (needs re-tuning from previous settings)
- increased range of accepted zombie tune parameter for GPUs with high interconnect bandwidth
- slightly decreased GPU load of Polaris GPUs during DAG build
- Added more control about handling cards that are detected to be non-working any more. Use parameter --watchdog off/exit/script to turn off any action, exit the miner with a specific exit code or to run an external script. See detail description on the 1.21 release page
- Nvidia cards that experienced a OpenCL driver error (e.g. "CL_OUT_OF_RESOURCES" will now also trigger the watchdog with the configured effect.
- The --ethstratum parameter can now take two options separated by a ',' to give different options in case the dual or split mining mode is used.
- The dns resolving and the connection attempt can now timeout (after 10 seconds each) and will re-try to connect afterwards. This fixes an issue when a pool went offline and the following connection attempt takes indefinitely much time. Each timeout event contributes to the counter that will trigger switching to fail-over pools.
- New option --apihost (default 0.0.0.0) which controls to which host address the api binds. Use 127.0.0.1 to restrict api access to only your computer, 0.0.0.0 is equivalent to everyone can access when rig is reachable on the used apiport. IPV6 ip addresses should be supported, but is untested.
Fixes
- Fixed a issue that might cause the rig to drop to 0 hash rate on epoch changes - including changes with activated ZIL caching
- Fixed the pool hash rate reporting not working correctly in dual & split stratum modes
- Fixed the dual stratum connection not picking up the correct worker name when --worker is used
- Fixed miner not loading Ethash / Etchash kernels on Tahiti and Hawaii GPUs when using older then end 2017 drivers.
Note on Watchdog use
There are different reasons why a card might crash and drop to 0 mh/s or g/s or sol/s. Often this happens when the card is slightly too much undervolted, but other problems like heat are possible. Additionally the OpenCL driver of Nvidia cards sometimes crashes with a CL_OUT_OF_RESOURCE error - this is rather a software then a hardware thing and will be fixed soon.
Anyways: Once a card is crashed some cards - mostly AMD cards - need a system reboot to get the faults card working again. Other cards - mostly Nvidia - just need a closing of the miner program - a few seconds wait time - and then are fine to get going again.
Therefore the crashed card detection now allows three different options to proceed with a crashed card or driver:
--watchdog off
This will do nothing except for printing a message. If only a single card did crash and not the whole driver this means the other cards will continue mining.
--watchdog exit
This will close the miner with a exit code of 42. This can be picked up by the .sh or .bat script that did start the miner (an example is provided in mine_eth.sh and mine_eth.bat) so the miner will restart after some seconds of pause. This is recommended option for Nvidia cards.
--watchdog script
With this option the miner will call an external script (default path is current working directory and there emergency.sh / .bat), which can be configured with --watchdogscript. The moment the script is called the miner itself will exit. The script needs to take care about rebooting the rig or informing the OS what to do. Since this was the default behavior in previous versions it also is the default. In case the script can not be found, an error will be printed and the miner will continue as with --watchdog off.
lolMiner as a GPU miner for AMD.