Topic: hsrminer - Nvidia mining software for various algos by palgin&alexkap - page 49. (Read 30791 times)

Дa... c кapтaми пpoблeмы ))) Я ж в Cитилинкe paбoтaю. Кapты жмy нa миллиoнныe peзepвы. Иx y мeня вce выкyпaют))) Цeны зaвышeны oчeнь cильнo я мoгy cкaзaть. T.e. вeндop oтдaeт кapты нaмнoгo дeшeвлe, чeм oни в poзницe, дa и чeм в мeлкoм и cpeднeм oптe. И Cитилинк eщe лoяльный игpoк. T.e. зaвышaeт нe тaк cильнo кaк тoт жe днc или oнлaйн тpeйд.
Зa ядpo cпacибo. Зa мaйнepoм бyдy cлeдить. Фepмы нa зeлeныx пepeвeл нa нeгo. 1% пpи пpиpocтe в 3% нe жaлкo oтдaть )))

Ecли бы былo oтдeльнoe ядpo пoд 1050Ti, тoгдa дa, нo y мeня дaнныx кapтoчeк нeт, тaк чтo нe нa чeм экcпepимeнтиpoвaть. Пepeд HГ зaкyпaть жeлeзки нe xoчy, в мaгaзинax aдищe, пocлe пpaздникoв зaймycь и нaпишy oтдeльныe ядpa для 1050Ti и 1080Ti, мoжeт бyдeт eщe yвeличeниe xeшpeйтa, нo, кaк yжe гoвopил, из кapтoчeк нижнeгo цeнoвoгo ceгмeнтa yжe пoчти вce чтo мoгли выжaли, xoтя вoн мoe ядpo пoд 1060 нa 1050Ti кaкoй-тo пpиpocт дaлo.

Now the 1050ti is working. Approximately + 2-3% in comparison with Alexis78
499-509 h / s

// ecли я твикaю интeнcиcнocть дo пpeдeлa, тo peзyльтaт пpиpacтaeт мизepнo. Ecть ли вooбщe cмыcл ee тpoгaть? Pocт зaгpyзки пaмяти идeт пpимepнo c 2400 мб дo 3490. A peзyльтaт eлe pacтeт.

UPDATE: fix for 1050Ti added, also 1050 (not Ti) experimental support, but I think 1050 owners will need to tweak intensity first.

Please, re-download and test.

Thank you for participation, I've planned nist5 to be next, but recent tests of my nist5 kernel showed that speed is a bit less than optimised alexis78's kernel (old ccminer-palginmod). As I state different kernel techniques in this project, I'll switch to something else, nobody'll use devfee miner when faster freeware variant is avaliable. It's still possibility to make it faster by another computational module work load, but it'll require dozen of time. Or I need to project faster groeslt kernel, current way of dealing with it is on the edge of hardware's capabilities, theoretically it can be 2 times faster, but it all requires tons of workhours from me.
Now I need to rest a bit, this weekend I'll be one lazy ass 


Thought that X17 is ASIC's theme now (Bainkal etc.) ? I'll take a look, of course, but won't promise anything until closer look.


1050Ti doesn't have kernels for it, I'll add experimental 1060 version and re-upload to github.

Ядep нa 1050 и 1050Tи нeт, пoтoмy и pyгaeтcя, тaк кaк пытaeтcя oбpaбaтывaть peзyльтaт paбoты кaкoй-либo дpyгoй кapты в pигe. 1070 и 1060 peзyльтaты y вcex paзныe, нo тyт yжe в вoзмoжнocти жeлeзa yпиpaeтcя, из ниx вce coки yжe выжaты, я тaк дyмaю.


Yes, this issue exists, but it's not affecting performance. I'll definitely look into it later, need to rest a bit first

My results so far

testing only neoscrypt ...

1080ti * tdp 100 * core +144 * mem +500 ... 1800kH (-i 5 or 6  no change, 7+ makes illegal memory acces)
1070ti Zotac Mini * tdp 95 * core +250 * mem +400 ... 1400kH+
1070 Evga SC  * tdp 95 * core +100 * mem +300 ... 1200kH (but with -i 4)

EDITED: now it seems, that evga 1080ti is goin on 1950kH (-i 6, 95 tdp, +100 core, +400mem)

the only "problem" is in recognition of GPU in stats ... im runing 2 instances for different intensity (4*1070 + 1*1080ti in 1 rig)
-d 1,2,3,4 is for 1070s ... -d 0 is for 1080ti
miner shows serparated 1080ti stats ok
but when running window with 1070, miner is writing 3x1070 and 1 1080ti in stats (only name of GPU1, stats itselfs are for 1070)

Hey. 1050ti does not work at all.
GTX 1080 = 1254 (1620/5315).

//1050ти нe зapaбoтaли - pyгaeтcя нa дyблиpoвaнныe шapы. Ocнoвнoй бycт идeт нa 1080/1080ти, кaк я пoнимaю, тaм пpиpocт oщyтимый. Ha 1070/1060 paзницa c нacтpoeнным ccminer Alexis 2-5%.

Thanks for the great work!

What about x17 algo? Verge is on a roll and is a major coin to mine

Tested on 2 1080ti

Hsr - 3435 Kh/s
Klaus - 2913 Kh/s

The improvement is very real.

Going to test on 1070 and 1060 6gb

Also please consider adding xevan algo.

Update:
Tested on a rig of mix 4 1070s and 1 1060 6gb
Both 1060 6gb and 1070 has 100 Kh/s improvement each.
1060 is doing 860. it can be increased a bit further but don't want to stress. I am impressed by the 1060 improvement though.
1070 doing 1227 still not fully overclocked and without using intensity values.

Great thanks, i'll definitly use it when my 1080ti rigs will be built in a couple of days, if it's stable keep up the good work man !

Ps : what algo do you have in mind, next ?

Hi, yes, it's CUDA 9.1 (just to keep actual CUDA version), so the lowest driver I think is 388.3x .

I've got dozen requests for ccminer-style API, it'll be added, can't say exact time bounds, but I'll work this way.

1070: same speed as Klaust 8.17 (-i 17.5)
1070ti: 4% faster  (-i 17.5 both)
1080ti: 20% faster (-i 17.9 max, 1830 kH/s with +80 core + 200 mem @250W vs 1650 kH/s at Klaust 8.17)

This is awesome!

The only problem is API. I miss Awesome miner monitoring with lack of API.

Hey,

Tried to use your miner, but it seems it require a certain version of nvidia drivers (is it cuda9.1 so needing the latest) ? im using 385.28 and 1070's, can you tell me the required lowest version ? Also any chance to get a working API ? it's really crucial to use  your miner in a production environment. Keep up the good work  

Cheers.

I need to clarify about devfee, there are two techniques implemented in miner, which one to choose miner decides itself during start:

1)2 stratum threads with 1% of CUDA threads running on dev
2)1 stratum thread running in cycle User-Dev-User (proportionally 99/1)

In Neoscrypt case only second variant avaliable regarding to kernel specification and during pool switches there are delays (sometimes 5-10 sec, depends on Dev pool and User pool load and connection stability). As average hashrate calculation is based not only on number of hashes, but also on the timeline (including connection delays), so after each cycle User-Dev-User switch average hashrate will be shown less than it actually is.

To sum up: check speed on pool or after 10-15 min from launch on miner side.

UPDATE: for 1060 3Gb and 6Gb users specifically, please, re-download binary from GitHub, fix for 6Gb version 1060's implemented.

Until any critical bug will be found or until I'll buy 1080Ti, Neoscrypt developement will be frozen from now, I'm switching to other algos.

My 1080TI if I am mathing right should be pulling only 230-235W miner was only showing 220W. It is all just setup using stock settings on driver version 388.31

My 1080 runs very hot because it is in a airflow limited case.

I would drop power limit on 1080Ti it can draw over 300W and your is on limit

mine at 80% is using ~200w I don`t like to push it more, clock is ~1400-1550mhz for GPU

Was going to update previous post but there were more replies to the thread than I expected.

Environment is inside closed cases ambient temps are 72F

Neoscrypt hashrates using Klaust
Getting 1434KH/s Temp 70c on stock clocks 1080TI FE
Getting 1186KH/s Temp 78c using stock clocks on my Zotac Mini 1080

HSRminer hashrates:
1080TI FE
No adjustments 26.9% increase miner side
speed is 1933.92kH/s, Temp 69c, zPool.ca 45 Min average shows 2133KH/s

Zotac Mini 1080
No adjustments 8.86% loss miner side
speed is  1016.32kH/s, Temp 80c, zpool.ca 45 Min average shows 1216Kh/s

guessing miner hash is calculating after devfee adjustment.

Now I just need to be able to fetch ccminer style stats and I'm using it for my 1080TI

For neoscrypt when you have a time best would be if power use can be optimized, all my cards are hitting power limit this heavy limited 1070 to ~120w have low speed gain 1-2%
the algo is so heavy on power

Thank you for feedback, according to developement plan at first I'll make algo coverage, then features will be added. Neoscrypt was quite a heavy job for me, so I got stuck for about 2 weeks, but I think I'll switch to lyra2v2 or nist5 now and return to Neoscrypt later.

UPDATE: re-uploaded Neoscrypt binary with small 1080 kernel tweak, can boost 20-30 kh

Test on my personal heavy undervolted settings with fixed voltage(800mv) and strap. 1695 core, 4000 (8000) mem with p2 off so no drop on mhz when mining.

1070:
Klaust: 1070 kh/s
Palgin: 1150 kh/s (+7.5%)

1070ti:
Klaust: 1240 kh/s
Palgin: 1367 kh/s (+10.2%)

But couldnt start it with all 6 cards, with 1 is ok, with 6 mem error, havnt tested on how many the problem occurs.
Dont have cards with gddr5x because the suck on ether, neoscrypt and monero.

Please add working api, so Nokia555 can add it to Ethmonitoring.com