Quote from: Lolliedieb on March 12, 2018, 02:58:08 PM

Hi folks,

well it took some time for me, but finally it is there: lolMiner-mnx 0.3 Cheesy

And its the first month birthday of the miner as well - time to party Grin

As usual here are the download links:

Linux 64: Click me (v0.3)
Windows 64: Click me (v0.3)

It will take a bit to describe what is new, so lets start.

Introduced a complete new Stratum Interface

As you may know the stratum interface of 0.22 was very buggy and caused serious problems on connection losses as well to the user as to the dev fee pool. Therefore 0.3 now has a complete new Stratum interface that should handle the situation much more gracefully and stable.

That said the new interface also comes with a new feature: fail-over support. You now can pass comma separated lists to --server, --port, --user and --pass. The miner will try to connect to the first server in the list with the first credentials. Once there are too many reconnection attempts for this first connection the miner will go to the next pool in list with the 2nd user name and so on.
With --max-connection-attempts you can define how many connection attempts to be used before the miner switches over to the next pool & credentials in list. The default is 5.

This new stuff also applies for the dev fee connection. To bring this down the current three biggest pools have to be down simultaneously. Moreover the dev fee connection will try to use the same pool as the user for the fee, so the miner does no longer damage the balance of the pools by sending all shares to one pool. If this is not possible, it will do a round robin through all pools in the integrated list.

Even if the fee is completely blocked the miner will now continue to work, but slower then usual, because it will pause in between for reconnection attempts. The performance loss is approximately 5-10% depending on card and settings, but it will continue to work.

Improvement of the OpenCL code

As for the OpenCL code I did some performance improvements, that help hiding the memory latency (by almost 40%). This results the following performance table. I only listed the devices available for testing.

Nvidia

	lolMiner-mnx 0.22	lolMiner-mnx 0.3
GTX 970	unknown	10.2k sol/s
GTX 1060	9.9 - 10.1k sol/s	10.8-11.1k sol/s
GTX 1070ti	15.2k sol/s	17.5k sol/s
GTX 1080 FE	13.4k sol/s	17.1k sol/s
GTX1080 Ti	19.5k sol/s	22.5k sol/s

AMD

	lolMiner-mnx 0.22	lolMiner-mnx 0.3
RX 570 8GB	6.9k sol/s	8.9k sol/s
RX 580 4GB	7.4k sol/s	9.1k sol/s
RX 580 8GB	7.6k sol/s	9.8k sol/s
RX 580 4GB	9.9k sol/s	10.3k sol/s
with Mining BIOS
Vega 64	16k sol/s	~19k sol/s (to be confirmed)

All of the above numbers were generated with the Linux version of the miner with current amdgpu-pro or cuda drivers.

Note that the performance of AMD cards may be limited in Windows, see the commends to set-work-batch below.

Beside this I added some new comfort features & changes:

--platform and device
The platform parameter now understand the keyword AUTO, which is also default. This one will select the platform with the most graphic cards automatically. If there are two with the same number, e.g. and Intel integrated GPU and one AMD / Nvidia graphic card it will prefer the non-Intel platform.

Device now detects two special values ALL and ALLGPU (default), that will select all devices or all graphic cards in the selected platform. Note then when using --device ALL on a processor with integrated graphics the miner will mine as well on the integrated graphics and on the CPU part. For most configurations this is not recommended.

--set-work-batch
There are now three special values 2G, AUTO and HIGH. When the work batch is set to 2G the miner is allowed to use 2 gigabytes of graphic memory at most. At level AUTO it will use at most 60% of the graphic cards memory and at level HIGH 90%. Note that AUTO and HIGH operate on per-GPU level, so mixing graphic cards with different memory sizes is no problem.

For Windows and AMD graphic cards there currently is a problem with lolMiner-mnx and the driver preventing that the miner works with more then 2 gigabytes of memory. Therefore the default for the Windows version is 2G, while the default for the Linux version is AUTO. This causes lolMiner-mnx 0.3 to be slightly faster on Linux compared to Windows when using AMD graphic cards.

The restriction does not apply for Nvidia graphic cards. For this GPU it should be save to use AUTO on any operation system.

New parameter: --use-compact-short-stat

This parameter will make lolMiner-mnx to print the short statistics in a similar style as the long statistics but without the surrounding dashed lines. Therefore one single line will be used for all devices instead of one line per device.

New parameter: --exit-on-zero

This parameter will cause lolMiner-mnx to check the speed of each device when printing out the long statistics. If an device is detected with 0 sol/s (thus it was idle on the long statistics intervall) the miner will quit, so it can be restarted from a watchdog script.

With all those changes I still may announce that working configuration files from 0.22 can be used with 0.3 and should work as before. But to use the new features you may like to modify it Wink

Beside the new comfort work I also tried to hunt down the memory access crashed and the reasons for reported 0 sol/s. This said it is natural that I may have overseen a problem, but it should be better now. If not please report.

That said there are two more smaller changes:
1. The dev fee is now reduced to 1.5%

2. There is a new readme in .pdf file format shipped with the miner. This is now much closer to a real manual and should help to configure the miner Wink

Bugs is a good keyword.
The fee collector collected slightly over 72 MNX for my early testers. That said I decided to round it up to 75 to make is a smooth number - or just because I have a good day Grin

But still I do not have all addresses yet, so here a remember to the following people still missing - Remember you have time until this Friday, 12:00 UTC to claim your part (!)

From bitcointalk forum:

Code:

AAile
Anon11073
Atomsprengkopf
BtcBlack
ExEric3
gaissa
goedi
HarleyQuinn2011
Inves
Kompik
mikebzh44
Monayman
NameTaken
ol92
salshicha
Sebahl
Syth
TheG
Wa4A
YoloMolo
yurez83

From telegram:

Code:

crodair
Racquemis
Reagan
Valery

Beside this I will do one more announcement. Unfortunately I am not good enough in doing batch programming for a cool Watchdog, so I will put up an contest for that during next week. More details will follow.

GTX970 4GB 6.8>7 KSOL/ V0.3 10KSOL

skyluis2000

Posts

Merits