[ANN] lolMiner-mnx, Mining MinexCoin(MNX) New version: v0.4 (July 22)

[Anon11073]

Hi folks,

well it took some time for me, but finally it is there: lolMiner-mnx 0.3
And its the first month birthday of the miner as well - time to party  

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 <number> 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  

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


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  
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.

Great job.

I am trying a new miner.
Very good performance. The miner displays 17k sol / s.
We will test the stability of the new miners.


P. S.
I will send you PM.

[1715582085]

1715582085

[1715582085]

1715582085

[1715582085]

1715582085

[1715582085]

1715582085

[1715582085]

1715582085

[Timoxa886]

Thanks for the work, but the 0.3 version gives less per 1000 s.s for the 1050 maps
The load parameter is now set not by a digit, but by the words 2G / AUTO / High, in my opinion this is a problem. on version 22, the 1050 card with load parameter 3 gives 6500ss, on version 0.3 it gives 5500 with a high
Maybe I'm setting the load parameter incorrectly ??

[bugnotme]

Lolliedieb,
some additions to your stat table:

Code:

		lolMiner-mnx 0.22	lolMiner-mnx 0.3
GTX 960		6.2k sol/s		6.7k sol/s

[ninjam]

@Lolliedieb
Tested on vega 56 under Windows 10, using all 3 parameters: 2G, AUTO and HIGH and results are not significantly higher.
Also there is still that huge ram usage spike at the beginning (around 14GB).
Anyway, thanks for adding additional functionality.

[Mogul59]

unfortunately the main problem remains:


can I start the miner incorrectly? please tell me the correct startup file. I have a 1070 and Win7

Code:

timeout 10
lolMiner-mnx.exe --server mnx.suprnova.cc --port 7076 --user *** --pass *** --device 0,1,2,3
pause

[valesthorm]

i have 3 1070 gtx and they working at 15400-15700 sol/s,and no oc,but power limit set over 100%.
What you doing ,for give more sol/s?(more 15000)

[WinstonBlue]

A faster miner thats for sure, but still unstable. On multi rig machines after a few minites the miner app crashes (no driver, the miner!) and a stratum problem with suprnova.cc ( connection lost, server not reachable)

Optiminer connects flawlessly.

[WinstonBlue]

unfortunately the main problem remains:


can I start the miner incorrectly? please tell me the correct startup file. I have a 1070 and Win7

Code:

timeout 10
lolMiner-mnx.exe --server mnx.suprnova.cc --port 7076 --user *** --pass *** --device 0,1,2,3
pause

The same problem here. Windows 7, 5 card rig.

[Mogul59]

i have 3 1070 gtx and they working at 15400-15700 sol/s,and no oc,but power limit set over 100%.
What you doing ,for give more sol/s?(more 15000)

this is very much. what settings do you use to start the program?

[Kompik]

I dont want any reward, please keep it, you deserve it because of your development

[valesthorm]

i have 3 1070 gtx and they working at 15400-15700 sol/s,and no oc,but power limit set over 100%.
What you doing ,for give more sol/s?(more 15000)

this is very much. what settings do you use to start the program?

lolMiner-mnx.exe  --device 0,1,2,3,4 --server mnx.suprnova.cc --port 7076 --user x --pass x
pause

[mk111]

V0.3

Been running for 30 mins.

2x 1070 on Win10
--set-work-batch HIGH

Average speed (60s): 17088.5 sol/s | 14157.2 sol/s Total: 31245.6 sol/s

Total jumps between 29000 and 34000 sol/s

[mk111]

V0.3

Win 10.
4 x RX 580 4GB
--set-work-batch AUTO

Each card does between 7000 and 9000 sol/s but the lolminer .exe uses 6GB memory out of the 8GB system memory.

[beta1234]

http://imgs.to/v/?frR6QGyUTg#frR6QGyUTg

Same old Problem

And i see no hashboost to 22,5kSol for 1080Ti, everything is pretty same like the Minerversions before
1080Ti, 1070, 1060 =-> 48K/Sol/s max

[Lolliedieb]

Same old Problem

And i see no hashboost to 22,5kSol for 1080Ti, everything is pretty same like the Minerversions before
1080Ti, 1070, 1060 =-> 48K/Sol/s max

Very strange... could only be that your driver really stops the OpenCL execution, because Kernel and host side are now dead lock free - so even when code does not finish properly it should schedule new Work for the GPU.  Just ordeted own hardware to do tests at home by the way.



As it appears we have first known problem with low end cards (gtx 750, gt 1030, gtx 1050...) for these chips the new algo seems in fact slower. I think I will do a switch to turn off code changes for this cards.

As said about the 0 sols: no idea why this happens. Anyone else?

[ninjam]

Tested this new version and although speeds reported in the miner are higher than for previous version (for nvidia rig), after about 4h speeds reported on the pool side are even less than previous version. For previous version avg results on the pool side (http://eu2.minexpool.nl) were reported correctly, maybe even few % more than in the miner, but for this 0.3 version it is around 20% less than what is reported in the miner. I'm talking about average results.

[skyluis2000]

Hi folks,

well it took some time for me, but finally it is there: lolMiner-mnx 0.3
And its the first month birthday of the miner as well - time to party 

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 <number> 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 

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


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 
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

[4ward]

rx480 with mining bios on Windows 10 w/ last drivers
2g - 10 kh/s
auto - 11.1 kh/s
high - 11.3 kh/s

no memory issues with driver preventing memory usage or anything
running with:
setx GPU_FORCE_64BIT_PTR 0
setx GPU_MAX_HEAP_SIZE 100
setx GPU_USE_SYNC_OBJECTS 1
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_SINGLE_ALLOC_PERCENT 100

ps.
previous version with workbatch 12 was showing 10.2 kh/s


pps. i have an issue with miner hangs when you scroll the text output in the window

[ol92]

titan xp with lolminer 0.3: 30S/s!!! Speed depends on strong overclock to memory (+800).

RX vega with blockchain driver : 19500 S/s with gpu@1408 and mem@1100, work batch 11

Congrat for the new miner performance!!!

[ol92]

Hi folks,

well it took some time for me, but finally it is there: lolMiner-mnx 0.3
And its the first month birthday of the miner as well - time to party  

...

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

--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.

With blockchain driver on vega, there are no problem with higher work batch. I found that 11 is a good choice for rx vega on win 10 with this driver.