Bitcoin Forum

Welcome to the discussion thread of on 名無し名誉名人(Nanashi Meiyo-Meijin)'s ccminer fork.
This thread is the branch of http://askmona.org/4314 in Askmona.


Features

• Faster Neoscrypt, Lyra2RE and Lyra2REv2 than the original ccminer
• Support for Pascal, Maxwell, Kepler, and Fermi GPUs
• Support Monacoin solo mining

Benchmarks
http://cryptomining-blog.com/7954-windows-binary-of-the-ccminer-1-7-6-r5-fork-with-faster-lyra2rev2/
http://cryptomining-blog.com/7997-windows-binary-of-the-ccminer-1-7-6-r6-fork-with-faster-lyra2re/
http://cryptomining-blog.com/8117-windows-binary-of-the-ccminer-1-7-6-r10-fork-with-fixed-neoscrypt-performance-on-pascal/


Downloads
Windows Binaries (1.7.6-r10) : https://1drv.ms/u/s!Aud1FauQ46vHhmd5NYbLOB2bQlOt (https://1drv.ms/u/s!Aud1FauQ46vHhmd5NYbLOB2bQlOt)
Source Code (1.7.6-r10) : https://1drv.ms/u/s!Aud1FauQ46vHhmh5NYbLOB2bQlOt (https://1drv.ms/u/s!Aud1FauQ46vHhmh5NYbLOB2bQlOt)
In this zip file, ccminer_CUDA8.exe is by compiling in CUDA 8.0. So it needs the latest Nvidia GeForce drivers.
ccminer.exe is by compiling in CUDA 7.5 and suitable for compute capability 5.0 or less of the GPU.(ex.750, 750ti, Kepler or Fermi)

Old binaries and sources are here.(Names beginning with ccminer-windows are source files.)
https://onedrive.live.com/?authkey=%21AHk1hss4HZtCU60&id=C7ABE390AB1575E7%21825&cid=C7ABE390AB1575E7


Tips
In older GPU than compute capability 5.0, it may not be very fast. This problem can be solved by compiling in CUDA 7.5.
And 32bit ccminer is faster than 64bit one.


History


Links
Monacoin Project : https://monacoin.org/en/
Askmona : http://askmona.org/
Tpruvot's ccminer : https://github.com/tpruvot/ccminer


Nanashi Meiyo-Meijin's BTC address 1P2SbHB5DaGPymu9C8WQCsEbnrw3dEWcDe
Please donate.

Lyra2re (r6, cuda 8 ) not good on my gtx750 (win7 64, latest nvidia driver), only 540khs, I got more then 1mhs with sp's cuda 7.5 version.
lyra2rev2 (r6, cuda8) is little bit faster for me then r5 cuda 7.5

wasn't this already released, what's the point of this? it was released by nicehash a few days ago

at first thought this was even faster, but it's not the case

no, it is not the same.

Thank you for testing.
Lyra2RE implementation is different in GTX9xx and GTX750.
Nanashi Meiyo-Meijin did not test to use GTX750 and GTX750ti. So he did not notice this problem.
He looks this thread. Maybe he will fix it in next update.

Nicehash forked Nanashi's ccminer1.5.80-r10.
This ccminer was released in May 27. And Nanashi switched the base ccminer from SP's one to tpruvot's one to support Pascal, Kepler and Fermi GPUs.
CCminer 1.7.6-r6 is faster than Nicehashminer.

i see the same hashrate mentioned on cryptoblog, what are the differences then if this is not the same?

Maybe a little patience is in order. The recent improvements to Lyra2 & Lyra2v2 are a welcome surprise as the prospects for
continued support of Cuda optimizations was not looking good.

These improvements were sponsored by Mona, which uses the lyra2v2 algo, I am hoping some of the talent can also be applied
to other algos if there is community support.

This is the best news I've seen in a long time and the timing couldn't be better with the release of Pascal, Cuda 8 & SM 6.1.

Thanks joblo. I couldn't say what I wanted to say.


Development schedule (Roughly)

• Debugging Lyra2RE
• Optimizing Neoscrypt
• Optimizing Cubehash in Lyra2REv2

Nanashi thinks several techniques used in Lyra2REv2 is applicable to other algorithms. The most typical algorithm are Lyra2RE and Neoscrypt.
I hope you test it or analyze it with Nsight in his next update.

if this is a long term project, then i'm supporting it, since sp want to say goodbye

it's good to have other that want to improve the code, especially for the new gpu

Nanashi Meiyo-Meijin has started this project since last month. It might change because this project has only just begun.
And he thinks this project is just a hobby. I hope he will have the fun of development.
but still he will rejoice if you support.

Trying to connect to miningrigrentals gives a bunch of 'stratum connection interrupted' spam and never connects.

Seems to work fine with just a straight connection to like Nicehash. Hope this isn't a sign of a miner fee that isn't reported.

now lastest version:
1.7.6-R6
and
1.7.6-R6-Fix
Wich version for maxwell cards ? (9xx series) ?

Now, Nanashi Meiyo Meijin released CCminer 1.7.6-r6-fix.
It is bugfix for 750 and 750ti, and optimized for Kepler and Fermi GPUs.
For Maxwell(like 9xx series) and Pascal GPUs, 1.7.6-r6 is probably faster than 1.7.6-r6-fix.


You should use 1.7.6-r6. and dont forget to install latest Geforce drivers.

Did this problem occur with only this ccminer?
No one uses miningrigrentals in Askmona, so it might take a long time to resolve this problem.

thx

Yes, it only seems to happen with this version. SP's ccminer doesn't have this problem nor DJMs version.

MRR can be used to mine monacoin or any coin that uses lyra2v2 if you rent rigs for that.

The same 500+khs unfortunately. SP version can do 1mhs+

Thank you for testing.
Since the optimization of policy for ambiguous, also he does not have a 750, he is in trial and error now.
I think it is very interesting that no more speed unlike 9xx.

Nanashi Meiyo Meijin released CCminer 1.7.6-r6-fix2.
It is bugfix for 750 and 750ti.
http://imgur.com/gup4rN1.png
For Maxwell(like 9xx series) and Pascal GPUs, 1.7.6-r6 is probably faster than 1.7.6-r6-fix.

Fantastic job! My poor gtx 750 now doing 1733kH/s Lyra2RE!

The best for lyra2v2 on gtx 980 is ?

Thanks .

After using the miner it appears that hashrate is not remotely consistent with what is reported by the miner. Poolside hashrate fluctuates from 50-100%

need changes about r7 version

The fastest lyra2v2 kernal for the maxwell chips is the sp-mod #7 private miner 0.1BTC (All nicehash algos optimized)

   
My version is currently 6% faster than the opensource  CCminer 1.7.6-r6 / r6-fix2 on the gtx 960.

WOW! Already avaliable R8 ver! need changelog!!!

I'm sorry I'm so late.
Nanashi Meiyo Meijin released CCminer 1.7.6-r8. (1.7.6-r7 was found some bugs.)
It increases the performance in Lyra2RE for Pascal, Maxwell and 750 or 750ti GPUs.
It is the same performance in Lyra2REv2 as compared with the old ones.
And it supports compiling by linux.

Windows Binaries (1.7.6-r8) : https://1drv.ms/u/s!Aud1FauQ46vHhmN5NYbLOB2bQlOt (https://1drv.ms/u/s!Aud1FauQ46vHhmN5NYbLOB2bQlOt)
Source Code (1.7.6-r8) : https://1drv.ms/u/s!Aud1FauQ46vHhmJ5NYbLOB2bQlOt (https://1drv.ms/u/s!Aud1FauQ46vHhmJ5NYbLOB2bQlOt)
In this zip file, ccminer_CUDA8.exe is by compiling in CUDA 8.0. So it needs the latest Nvidia GeForce drivers.
ccminer.exe is by compiling in CUDA 7.5 and suitable for compute capability 5.0 or less of the GPU.(ex.750, 750ti, Kepler or Fermi)

Wich coin is using the old lyra2re algo? Lyrabar,vertcoin etc has moved to lyra2v2

It appears there is no performance difference with r8 vs r6. Tried the normal version and cuda8. This was done on a 1070.

Also it appears the miner is still not compatible with miningrigrentals. You get constant disconnects with 'stratum connection interrupted' message spam.

Nanashi read http://cryptomining-blog.com/7973-testing-the-nvidia-geforce-gtx-1080-founders-edition-for-crypto-mining/ and knew that Lyra2RE and Neoscrypt are too slow at GTX 1080.
So he has begun to develop. It does not matter whether or not useful.
Lyra2RE Cryptocoin is Crypto(CTO). We dont know only it...


Thanks testing. My GTX980 improves 5% in Lyra2RE. So I thought it is the same in Pascal...
Poolside hashrate problem is we also could be confirmed. It will may be resolved soon.
And I signuped miningrigrentals and tried to verify this problem. but i cant be confirmed it. Could you upload screenshot or commandline?

https://i.imgur.com/pnnPj6t.jpg

ccminer.exe -a lyra2v2 -o stratum+tcp://us-central01.miningrigrentals.com:3333 -u  -p x


I opened a ticket at MRR as well, they said they enabled debugging, but haven't said anything since. With other miners, sometimes this happens, but it will connect after maybe 10 seconds of spam.

Crypto has a marked value of 1000$ and a turnover of $1 per day.

http://coinmarketcap.com/currencies/crypto/#markets

There has to be another coin. Anyone?

Nicehash is paying around 0,512BTC a day for all the lyra2re miners.

Wich coin is supporting the algo?

change the port in your config to the port its trying  to reconnect to. (as im typing i cant see the image. port 50093?) rather than the 3333 to 50093

Don't alway look for a hidden gems or even logic when it comes to nicehash. When sharkcoin was "big" (for some at least) solomining was paying way worse than mining sharkcoin via nicehash.

Some people probably just can't do math or use bots properly.

Yup, that's a work around, however, it doesn't fix it. Here is the response I received from MRR to the issue.

"Your miner needs to be able to handle client.reconnect command, which in this case it looks as if it does not. there may be an option to enable it, please consult your miner software. please note that the rig port may change if it has not been active in two days."


Sometimes... There are a lot of hidden gems when it comes to mining though. Could be just a unknown coin some where laying low.

With CCminer 1.7.6-r6 on gtx980 OC i have 29.50 mh/s on lyra2v2

With CCminer 1.7.6-r8 on gtx980 OC i have 29.25 m/s on lyra2v2

Im on win 10 x64 .

Nanashi Meiyo Meijin released CCminer 1.7.6-r10. (1.7.6-r9 was released a while ago.)
It increases the performance in Neoscrypt for Pascal, Maxwell.
Lyra2RE and Lyra2REv2 are same in older versions. So Poolside hashrate problem and miningrigrentals bugs are not resolve.


Benchmarks in GTX980

r10(cuda8.0) 653kH/s 190W
r10(cuda7.5) 645kH/s 190W

r9(cuda8.0) 494kH/s 170W
sp1.5.80(cuda7.5) 599kH/s 188W
djm34(cuda6.5) 638kH/s 194W (http://cryptomining-blog.com/6450-new-ccminer-fork-from-djm34-with-faster-neoscrypt/)


Benchmarks in GTX1080

r10(cuda8.0) 1013kH/s
r9(cuda8.0) 412kH/s


Windows Binaries (1.7.6-r10) : https://1drv.ms/u/s!Aud1FauQ46vHhmd5NYbLOB2bQlOt (https://1drv.ms/u/s!Aud1FauQ46vHhmd5NYbLOB2bQlOt)
Source Code (1.7.6-r10) : https://1drv.ms/u/s!Aud1FauQ46vHhmh5NYbLOB2bQlOt (https://1drv.ms/u/s!Aud1FauQ46vHhmh5NYbLOB2bQlOt)
In this zip file, ccminer_CUDA8.exe is by compiling in CUDA 8.0. So it needs the latest Nvidia GeForce drivers.
ccminer.exe is by compiling in CUDA 7.5 and suitable for compute capability 5.0 or less of the GPU.(ex.750, 750ti, Kepler or Fermi)

thanks but why comparing the hashrate with the old djm34 version? the new one is much faster.

EDIT: according to crypto mining blog, the 1070 gets less than 900 Kh/s with this fork, I get much more with standard tpruvot's.
EDIT2: someone should tell that japanese guy he's working on old stuff.

Just tested this r.10 with neoscrypt on my super jetstream 1070.
960 kh/s, but it IS power hungry ))

pallas, could you please provide a link to best neoscrypt miner for pascal (win 7)

you can try 1.8 version of epsylon3 i remember he ported the djm34 version in the last version of his miner

Thanks! But I will need to compile with new cuda framework installed. My old compile machine is dead now and I have no time to build new one ((
Will wait for official windows binary release from tpruvot/epsylon3

I compiled it with cuda 7.5 and it worked pretty fine on 1070, but it is slower on maxwell.
Besides, the nanashi version is slower on all my cards.

What the max speed was you enable to get from one 1070 on neoscrypt?

1020 - 1040 Kh/s at 125 W

1.7.6-r10 Neoscrypt codes are based on SP's ccminer 1.5.80.
New ccminer 1.8 is already known by Nanashi.
https://github.com/tpruvot/ccminer/commit/4ca7b5a404e8acb8e863c8171beb256034f303c8
https://github.com/tpruvot/ccminer/commit/a4196b341d74b5abeba8239da8c7125ed4fe0f82
This code is roughly the same as 1.5.80. A Different point is the default intensity settings. And We checkd its Neoscrypt hashrate is not faster than 1.5.80.
But 1070 seems an exception...? Anyway, Thanks for testing.

Sorry Sir but I've tested both the 980ti and 970 and they are also slower. Please tell nanashi to rebase his work on latest djm34 or tpruvot  (which is the same as djm but with pascal fixes which in my case are slowing down maxwells).

Faster in Neoscrypt

Nanashi
1080
980

latest Tpruvot
1070
980ti
970

It looks like this. Why the difference is occurred?
Nanashi reads this thread so he will consider your opinion.

with your version i'm getting 900khs sam as 1.8 from epsylon, but i see people reaching above 1MH, with same oc on same system, on linux you get,, more but i don't care about linux

so maybe something on my end, but i doubt, since all other algo are on par...

which would be better (mhs/watts) for neoscript mining, 1070 or 1080? Tks.

Now neoscrypt mining is good only in linux with pascal. On windows 1070 can do 1mhs only with high oc and more then 100% tdp power. Maybe you should wait for windows fix driver with new wddm mode that hope will come at Aug,2.

1080 does not scale right, too expensive without offering overall a good hash, maybe only in a few algo like lbry

New types of gpu memory have a hard road to mining. HBM and GDDR5x are worse than this old good GDDR5.

this mean that Volta will be bad, because in the future every class of gpu will have those

maybe there will be new tool to make mining good with them



talking about the lyra2 algo, i think it need an optimization, if you can ask nanashi to look at it, because it's using 50% of the tdp

so i suspect that the hashrate should be doubled

any news on the poolside hashrate issue?
thanks!

I fixed it myself.
Not trivial.

this post https://bitcointalk.org/index.php?topic=1464999.msg15840607#msg15840607 caught my attention

there is somewhere a r12 release?

Hi,

there is no 1.7.6 r12 release look at Nanashis OneDrive here: https://1drv.ms/u/s!Aud1FauQ46vHhmd5NYbLOB2bQlOt (https://1drv.ms/u/s!Aud1FauQ46vHhmd5NYbLOB2bQlOt)

>>> But there is a new and faster CCminer 1.8-r2 Fork release! <<<

faster for lyra2v2 or just a porting from epsylon work?

It comes as sources only and requires cuda 8 toolkit to be installed (( My nvidia developer subscription is expired so I cannot download cuda 8 RC and build new nanashi version ((

That's weird as I wanted the sources and downloaded the windows binary instead by mistake.

what version is that exactly, the same as the last one of epsylon?

use the link to OneDrive above but cancel download. Then choose the last download in the list from OneDrive. There will be 1.8 r2
shame on me ... found some binaries ... will try

i uploaded it https://www.dropbox.com/s/r5fajizspbjzb9k/cuda_8.0.27_win10.exe?dl=0

Is it just me or someone else getting error can't validate CPU msg?

I got n1070 and best i manage to get is around 36-37Mhas/s on vert/mona coin