Hey Folks,
I just made the link to the first
lolMiner 0.6 alpha release public.
You can find the new version here:
Readme:
https://github.com/Lolliedieb/lolMiner-preview/blob/master/README.mdRelease:
https://github.com/Lolliedieb/lolMiner-preview/releasesChanges to 0.5 version:
- All algorithms removed except Equihash 144/5
- Faster mining on 144/5 compared to 0.5 stable release
- 0.6 uses now only 1850 MB of GPU memory and thus works on 2G cards
- Initial tuning for Nvidia RTX 2000 series
- Improved API
Details about the new API:
lolMiner 0.6 comes now with full http headers for the API, so the mime type is correct. As far as
I know now all common browsers can show the API output correctly. This also helps getting the
API output with tools like curl or any mining watchdog.
To call the api you now need the address /summary, e.g. localhost:8080/summary if your API port
was 8080. The preview version is not yet implementing the new 0.6 API in full (but the subset send
will not change in format). Next days I will put a list online with all the planned changes.
Note:
under /summary_trim you can find a API output where all white spaces and line-breaks are removed.
May help for some tools.
Speed estimates in new version (Equihash 144/5 only):
AMD:
RX 580: 27-29 sol/s
RX 560: 14 sol/s
Nvidia:
GTX 1080: 53-56 sol/s
GTX 1070: 40-45 sol/s
GTX 1060: 26-30 sol/s
more values to be added when available
Some results from tuning experiments:
AMD:
Clock is not so important here, memory bandwidth seems to be the key.
Tested setting was: 1200 / 1850 with -0.1v core and -20% power target
Nvidia:
lolMiner scales better with available compute power then most competitors.
(For example I saw it usually faster then ewbf for 1080 but slightly slower on
1070 cards, because the balance of compute power and memory bandwidth
is different between the two cards. )
Try to undervolt to get less energy consumption, but try to find settings
so that clocks do not drop too sharp. Setting a low power target without
undervolt may kill the performance.
Setting for the 1080 above were founders edition stocks. 1070 was tested at
70% PT and at stock clocks. 70% PT without undervolt gave 42 sol/s, stock
clocks 45 sol/s.
Whats coming next:
In the next versions I want to extend the API and implement a new work scheduler
which hopefully fixes the problems with more then 5 GPUs and reduces CPU load.
*Crossing fingers*
Also I want to apply the algorithm changes to the other algorithms as well and add
them successively to the alpha version