KopiemTu 3.0 - ethash & equihash - NVIDIA USB linux distro

[jk_14]

hey jk_14,
  I was active a long time ago on your litcointalk thread (different username) - and like to leave some "thank you" in here!
You were one of the reasons I dived deep into linux and I really did not regret it. Today I use "regular" linux installations (on SSD) on my rigs, but would be more than curious to take a closer look into your new kopiemtu version.

I mined on your vtc & bbr adresses from time to time while using your kopiemtu,
but like to send you a little additional donation. please post your BTC (or BBR) address on this thread.

take care!

Thanks! and I'm glad because of your linux improvement (that was one of reasons KopiemTu was born )

I have almost finished new distro v3.0

but even ETH is rather for AMD - I planned to include it there.
before - as Ethereum newbie - I successfully compiled an ethminer under old distro, with old Cuda toolkit

But now, under new Ubuntu 16.04.1 (i.e. Lubuntu), with new Cuda 8.0 - I have some issues yet.
Is there anybody, who has compiled ethminer with these two above?

[1715556526]

1715556526

[1715556526]

1715556526

[1715556526]

1715556526

[jk_14]

hey jk_14,
  I was active a long time ago on your litcointalk thread (different username) - and like to leave some "thank you" in here!
You were one of the reasons I dived deep into linux and I really did not regret it. Today I use "regular" linux installations (on SSD) on my rigs, but would be more than curious to take a closer look into your new kopiemtu version.

I mined on your vtc & bbr adresses from time to time while using your kopiemtu,
but like to send you a little additional donation. please post your BTC (or BBR) address on this thread.

take care!

Thanks! and I'm glad because of your linux improvement (that was one of reasons KopiemTu was born )

I have almost finished new distro v3.0

New master rig main page:



I will send you soon an early access link (free of charge )
just to take a closer look into a new version - and your feedback would be great

Things onboard:

Code:

root@kopiemtu:~#
root@kopiemtu:~# dmesg | grep NVRM
[   27.317822] NVRM: loading NVIDIA UNIX x86_64 Kernel Module  370.28  Thu Sep  1 19:45:04 PDT 2016
root@kopiemtu:~# nvcc -V | grep release
Cuda compilation tools, release 8.0, V8.0.44
root@kopiemtu:~# ls -l /opt/miners/
total 16
drwxrwxr-x 21 root root 4096 Oct  9 23:30 ccminer-1.8.2-tpruvot
drwxrwxr-x 21 root root 4096 Oct  9 06:51 ccminer-vert-1.5.80.1-vert
drwxrwxr-x 21 root root 4096 Oct  9 06:58 ccminer-vert-1.5.80.4-vert
drwxr-xr-x  2 root root 4096 Sep 17 22:15 ethminer-0.9.41-genoil-1.1.7
root@kopiemtu:~#
root@kopiemtu:~# cat /root/README.txt

(ethminer is under construction yet)

[jk_14]

uploading...

EDIT:
bug fixing...

[jk_14]

uploading...

EDIT:
bug fixing...

First post updated, only file image is missing yet:
https://bitcointalk.org/index.php?topic=520998

Uploading again... less than hour yet, so in meantime, some hints:

In fact new version is configured for GTX 1000 series, but if you want to run some older cards, even 750 Ti
- go into /opt/miners and in all ccminer's directories change a symbolic link ccminer to ccminer_all, which supports compute 5.0 up to 6.2

Code:

rm ccminer
ln -s ccminer_all ccminer

If you have some fast pendrive - you can upgrade system packages, but:
option "dist-upgrade" - will require additional tasks for proper operation

It's really nice to have master rig on SSD drive, and rest of rigs may work on 16GB pendrives (cheaper, and no SATA / power cabling)

For that purpose - create first KopiemTu pendrive, even under Windows
and when you see the first KopiemTu 3.0 screen with "boot:" prompt, type:
rescue <Enter>

then verify with partx, if your SSD is /dev/sda, and KopiemTu pendrive is /dev/sdb - like below:

Code:

root@sysresccd /root % partx /dev/sdb
NR    START      END  SECTORS  SIZE NAME UUID
 1     2048   821247   819200  400M
 2   821248 25397247 24576000 11.7G
 3 25397248 29493247  4096000    2G
root@sysresccd /root %
root@sysresccd /root % dd if=/dev/sdb of=/dev/sda bs=1M count=14401

- then you can easily add fourth partition on your SSD, with fdisk command.


Rescue system is quite usefull yet for:

1. utilising all USB ports for making many KopiemTu pendrives simultaneously:

Code:

dd if=kopiemtu-3.0.img of=/dev/sdb bs=1M count=14401 &
dd if=kopiemtu-3.0.img of=/dev/sdc bs=1M count=14401 &
dd if=kopiemtu-3.0.img of=/dev/sdd bs=1M count=14401 &
...

2. repair KopiemTu filesystems if broken (in case of lack of power, etc)

Code:

fsck -y /dev/sda1
fsck -y /dev/sda2
fsck -y /dev/sda3

[jk_14]

Done!

[jk_14]

Done!

most important, if you don't know that under KopiemTu - switching 8 rigs to new pool takes only 20 seconds:

https://bitcointalk.org/index.php?topic=167229.msg7716030#msg7716030

- that is old thread, now use:

Code:

mmip; mmih; mmit

and of course you can still control single rig from master command line, by:

Code:

mmip w12; mmih w12; mmit w12

[jk_14]

CRYPTO MINING BLOG info:

http://cryptomining-blog.com/8289-new-kopiemtu-3-0-linux-mining-distribution-for-the-latest-nvidia-gpus/


P.S.
while waiting for your v3.0 feedback, let me include the older one:


"The windows users have to try it for themself, before the really know what its like to use a easy to use and stable linux rig with failover support.
There are so many reasons for KopiemTu and you do really good work in supporting the users.

The main reason I tried kopiemtu some time ago was - i ran out of windows lic's and every AMD user were so happy about their BAMT.
After taking some time to get into kopiemtu, it became the only way I could think of operating a few rigs without having to look all the time
- if there is some problem with the pool or the system.  It is really no work to set them up and absolute stable... fire and forget.

Before I wasted lot of hashes with not having failover support using windows. Now I ran those rigs for weeks without any reboot.
I could go for a vacation and see if everything is fine with your http monitor (via vpn connect)... and switch the algos with via ssh."

https://bitcointalk.org/index.php?topic=167229.msg7716432#msg7716432

[induktor]

WOW!  WHAT a GOOD NEWS!
JK_14 you are a genius!

I will tested it tonight.

I have completely modded your Kopiemtu 2.0, Driver 376.44, CUDA 8, ccminer (TPRUVOT and SP) and GENOIL'S ethminer (117 works, 119 does not compile no matter what i do ) 3x GTX 1070 working very well on AM1 low power motherboards.

but of course it is a mod (a very VERY jerry rigged mod  hehe), not the real thing 

what OS is this one based on? Lubuntu 16.04 ?, i will tested it tonight and try to include ethminer (never compiled ethminer on 16.04)

[jk_14]

WOW!  WHAT a GOOD NEWS!
JK_14 you are a genius!

I will tested it tonight.

I have completely modded your Kopiemtu 2.0, Driver 376.44, CUDA 8, ccminer (TPRUVOT and SP) and GENOIL'S ethminer (117 works, 119 does not compile no matter what i do ) 3x GTX 1070 working very well on AM1 low power motherboards.

but of course it is a mod (a very VERY jerry rigged mod   hehe), not the real thing  

what OS is this one based on? Lubuntu 16.04 ?, i will tested it tonight and try to include ethminer (never compiled ethminer on 16.04)

Hi, Induktor!

Excuse me for that v3.0 ninja launch - that's because I discovered I love GTX 1070 as much as 750Ti long time ago
Great that you have some experiences with ethminer - I'm sure together we will solve that small issue, under current Lubuntu 16.04.1 and CUDA 8.

and I have the proof that not me, but you are genius, in fact
voila:

https://archive.litecointalk.org/index.php?topic=16800.msg277720#msg277720

https://archive.litecointalk.org/index.php?topic=16800.msg277876#msg277876

haha

[PVmining]

hey jk!
  I gave your new KT3.0 a try on a single testrig - it is a fine progress.
I like the new mine & monitor status. nice.


Some thoughts:

• using the "nvidia-smi -pl" function within the mining conf. I use this often to get a good balance for hash & wattage.
• showing the wattage in the http status monitor too.
• cuda8 is not faster in general. an additional cuda7.5 would not hurt.
• compute_61 is slower in every case I know. I would suggest to focus on 52 for the pascal cards.
• using yiimp as a LBRY donation pool and directly supporting epsylon3 with this choice  
• I see ETH is on you todo list?

one thought came up here.
the OP of this thread is reading more complicated than the use of kopiemtu is.
I think it would be a nice idea to make some little "howto video" to show how fast the the first rig could be online and mining.

ps. send you a little donation. cheers!
pps. you are crazy. and i like that I saw your outdoor rig on the other thread. lol. would love to see some more pictures about it.

[jk_14]

hey jk!
  I gave your new KT3.0 a try on a single testrig - it is a fine progress.
I like the new mine & monitor status. nice.


Some thoughts:

• using the "nvidia-smi -pl" function within the mining conf. I use this often to get a good balance for hash & wattage.
• showing the wattage in the http status monitor too.
• cuda8 is not faster in general. an additional cuda7.5 would not hurt.
• compute_61 is slower in every case I know. I would suggest to focus on 52 for the pascal cards.
• using yiimp as a LBRY donation pool and directly supporting epsylon3 with this choice  
• I see ETH is on you todo list?

one thought came up here.
the OP of this thread is reading more complicated than the use of kopiemtu is.
I think it would be a nice idea to make some little "howto video" to show how fast the the first rig could be online and mining.

ps. send you a little donation. cheers!
pps. you are crazy. and i like that I saw your outdoor rig on the other thread. lol. would love to see some more pictures about it.

Hi, PVmining!

I hope you have tested the current image, that what I provided you earlier is before last bug fixing.

- there is /root/powerlimit.conf file, where you can set required power level, then it's used during 'mine start'
because I have observed spontaneous power increase after days of working - I set /etc/crond.hourly/powerlimit, which set that back to requested level, just for sure
- so showing the wattage in the http status monitor is not so necessary due to this?
- I didn't check if double Cuda toolkits is complex issue
- because compute_52 is faster on GTX 10x0 than compute_61 - default symbolic link point to the ccminer_52, not ccminer_61, so it's done
- I will think about it - (after 1.8.2. my donation was made, https://bitcointalk.org/index.php?topic=770064.msg16337641#msg16337641 )
- ETH is in TODO, not super urgent while there are NVidia coins more profitable still

yes, such Youtube howto would be nice...
in fact, every first rig is instantly online and mining (...only pointed to my Coinotron worker )

Thank you for your generous donation, 0.0333 btc

I was asking for that:

http://www.piwo-zywiec.na-pulpit.com/tapety/szesciopak-piwo-piwo-ywiec.jpeg

but you have sent me this:

http://frega24.pl/media/catalog/product/cache/1/image/800x600/c562ae099580d6a062289618533acacb/5/9/5901559130000_3_15.jpg

... and you wrote I'm crazy


more pictures of outdoor rig is sent to you via PM ( also cool? )

[PVmining]

It was the latest version. I saw your bugfix info in this thread and redownloaded the image before installing on the stick.
I will look into your powerlimit.conf, interesting concept.

- I didn't check if double Cuda toolkits is complex issue
Its not. Meanwhile I use three different cuda versions on my rigs at the same time.
Just point the miner on the right version before compiling.
maybe it resolves you ETH miner problem on an easy way.

Code:

~cuda 
cuda-6.5  
cuda-7.5
cuda-8.0

- I will think about it - - (after 1.8.2. my donation was made,...)
To get me right. I don't meant to send all the hash for epsylon3. Just the poolfee as little form of donation.

[jk_14]

Regarding Cuda versions, that result was some time ago under previous Cuda:

Code:

[2016-07-21 07:21:25] accepted: 44/45 (97.78%), 3151 kH/s yes!
[2016-07-21 07:21:35] coinotron.com:3337 neoscrypt block 1305925
[2016-07-21 07:21:35] GPU #2: GeForce GTX 1070, 1076 (T= 67C F= 37% C=1803/3802)
[2016-07-21 07:21:35] GPU #0: GeForce GTX 1070, 1010 (T= 66C F= 36% C=1792/3802)
[2016-07-21 07:21:35] GPU #1: GeForce GTX 1070, 1063 (T= 67C F= 37% C=1787/3802)
[2016-07-21 07:21:55] GPU #0: GeForce GTX 1070, 1020 (T= 66C F= 36% C=1792/3802)
[2016-07-21 07:21:55] accepted: 45/46 (97.83%), 3153 kH/s yes!

https://bitcointalk.org/index.php?topic=1534558.msg15655645#msg15655645


at 3.0 release, under Cuda 8.0 and ccminer_52, with the same hardware and powerlimit 150 - I have:

Code:

Oct 15 21:46:58 kopiemtu ccminer[809]: coinotron.com:3337 neoscrypt block 1425668
Oct 15 21:46:58 kopiemtu ccminer[809]: GPU #0: GeForce GTX 1070, 1132 (T= 68C F= 36% C=1809/3802)
Oct 15 21:46:58 kopiemtu ccminer[809]: GPU #2: GeForce GTX 1070, 1138 (T= 70C F= 40% C=1784/3802)
Oct 15 21:46:58 kopiemtu ccminer[809]: GPU #3: GeForce GTX 1070, 1147 (T= 68C F= 36% C=1809/3802)
Oct 15 21:46:58 kopiemtu ccminer[809]: GPU #1: GeForce GTX 1070, 1057 (T= 65C F= 28% C=1809/3802)
Oct 15 21:46:59 kopiemtu ccminer[809]: GPU #1: GeForce GTX 1070, 1006 (T= 64C F= 28% C=1797/3802)
Oct 15 21:46:59 kopiemtu ccminer[809]: accepted: 21/21 (100.00%), 4408 kH/s (yes!!!)

it seems Cuda 8.0 is ok for my configuration...

P.S.
I have found some small bug, but easy to remove: after first boot monitor process should be stopped for 1-2 minutes...

[induktor]

Hello
I am working on the necessary mods for ethereum on your kopiemtu 3.0 and also for my display  

OK, it doesn't work, i installed a lot of libraries that in my notes are marked as necessary but still does not compile and the error is weird.

I even tried to move my entire cpp-ethereum folder (compiled and working in kopiemtu 2.0 MOD #4) and run it, also does not work.

If i try to run the compiled version i had in kopiemtu 2.0 in kopiemtu 3.0 i get:

Code:

Oct 16 18:06:46 kopiemtu root: /opt/miners/cpp-ethereum/build/ethminer/ethminer: symbol lookup error: /opt/miners/cpp-ethereum/build/libstratum/libethstratum.so: undefined symbol: _ZN4Json6Reader5parseERKSsRNS_5ValueEb

If i try to compile genoil's 1.1.7 cpp-ethereum in kopiemtu 3.0 i get:

Code:

[62%] Building NVCC (Device) object libethash-cuda/CMakeFiles/ethash-cuda.dir/ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o
nvcc fatal    :  redefinition of argument 'std'
CMake Error at ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o.cmake:207 (message):
       Error generating
       /opt/miners/cpp-ethereum/build/libethash-cuda/CMakeFiles/ethash-cuda.dir//./ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o

any ideas?
I will start digging on that particular error.

Also I found out that we have different CUDA 8 builds!!!!! that could be the issue as well.

Working: kopiemtu 2.0 MOD #4, CUDA version: 8.0, V8.0.26    - Driver: 367.44
Not working: kopiemtu 3.0 orig,  CUDA version: 8.0, V8.0.44    - Driver: 370.28

I don't think it is a driver issue, but it could be that something changed on the CUDA build.

[induktor]

Compiling ethminer update:

it seems that cmake fucks something up setting up std c++11 twice unnecessarily for some reason, the error in question is on libethash-cuda/CMakeFiles/ethash-cuda.dir/stuff  

editing flags.make there and removing the --std=c++11 creates all sorts of problems.

but using VERBOSE=1 during make -j4 i can see that NVCC is right, std c++11 is defined TWICE, removing the second one, fucks something up, but i have no idea how to remove the first one.

this is the line with the error (it may not be the only one, but until we/I/whoever!   fix this we probably won't see the next ones.

Code:

/opt/miners/cpp-ethereum/build/libethash-cuda/CMakeFiles/ethash-cuda.dir//ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o.NVCC-depend -ccbin /usr/bin/cc -m64 --std c++11 -DETH_JSONRPC -DETH_READLINE -DETH_ETHASHCUDA -DETH_STRATUM -DETH_CURL -DETH_FRONTIER -DETH_TRUE -Xcompiler ,\"-Wall\",\"-Wno-unknown-pragmas\",\"-Wextra\",\"-Wno-error=parentheses\",\"-pedantic\",\"-DSHAREDLIB\",\"-fPIC\",\"-fuse-ld=gold\",\"-O3\",\"-DNDEBUG\",\"-DETH_RELEASE\" --std=c++11 --disable-warnings --ptxas-options=-v -use_fast_math -lineinfo -gencode arch=compute_61,code=sm_61 -O3 -DNVCC -I/usr/local/cuda/include -I/opt/miners/cpp-ethereum/build/gen -I/opt/miners/cpp-ethereum/build -I/opt/miners/cpp-ethereum/build/libethash-cuda -I/usr/local/cuda/include -I/opt/miners/cpp-ethereum/libethash-cuda/..
nvcc fatal   : redefinition of argument 'std'
CMake Error at ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o.cmake:207 (message):
  Error generating
  /opt/miners/cpp-ethereum/build/libethash-cuda/CMakeFiles/ethash-cuda.dir//./ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o

any clues where can i look?, i will keep digging cmake to see if i can nailed it, because i really don't want to change cuda 8 and drivers, is pain and takes forever and may not even be the problem, could be very easily a cmake/c++/gcc version issue, dunno.

--== EDIT: ==--
UPDATE: in CMakeCache.txt the line that says:

Code:

//Propagate C/CXX_FLAGS and friends to the host compiler via -Xcompile
CUDA_PROPAGATE_HOST_FLAGS:BOOL=ON

I changed to OFF and now I compile until 95% without errors when i get:

Code:

[95%]
/usr/bin/ld.gold: error: ../libethash-cuda/libetash-cuda.a(ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o): requires unsupported dynamic relloc 11; recompile with -fPIC

I am lost... i have no fucking idea how to recompile with -fPIC, this has to be something related with a cmake config issue due to different version, will try to go back with cmake, make, gcc and g++ to the older version i use in kopiemtu 2.0 (if it is possible, since kopiemtu 2.0 is 14.04 and this is 16.04, we'll see.

--== EDIT 2 ==--

FUCK YEAH, it was the Goddamn cmake muufucka.
downgrade to version 3.2.2 and it........   almost....... compile.

instructions to downgrade:
first get rid of cmake, apt-get remove cmake

then:
cd
wget http://www.cmake.org/files/v3.2/cmake-3.2.2-Linux-x86_64.tar.gz
tar -zxvf cmake-3.2.2-Linux-x86_64.tar.gz

go to the app dir and move the files in bin to /usr/bin
and move the folder cmake-3.2 to /usr/share/cmake-3.2 so cmake will not complain.

now remove and recreate build in epp-ethereum and cmake -DCOMPUTE=61 -DBUNDLE=cudaminer ..
it reaches 100% and then explodes with a lot of JSON errors:

sample:

Code:

../livedevcore/livdecore.so: error: undefined reference to 'Json::Value::operator[] (std::string const&) '
../livedevcore/livdecore.so: error: undefined reference to 'Json::FastWriter::write (Json::Value const&) '

i got like 50 of those lines, all json related.
any ideas?
at this point, i am tired, will wait until someone throws me a bone   and will give it a try later.
I need a beer, or two, or three.....
 

[jk_14]

Working: kopiemtu 2.0 MOD #4, CUDA version: 8.0, V8.0.26    - Driver: 367.44
Not working: kopiemtu 3.0 orig,  CUDA version: 8.0, V8.0.44    - Driver: 370.28

V8.0.26 is RC (release candidate)
V8.0.44 is official Cuda 8.0

btw: If you want newer 367 driver, in V8.0.44 driver 367.46 is included,
but I decided to use 370.28 due to O/C in description

as far as I remember:

Ubuntu 14.04 with Cuda 8 RC was OK for ethminer,
the problem was with 16.04.1 and Cuda 8 RC (and Cuda 8 official)

my shortcut was to move compiled ethminer under 14.04 and include missing libraries
you can check the result if you run ethminer in v3.0 - it's not working still, one of libraries has still unresolved symbols
(by mistake I overrided my flash usb with old distro and compiled ethminer - it prevent me a bit solving it further...)

I installed gcc 4.9:

Code:

root@kopiemtu:~# update-alternatives --config gcc
There are 2 choices for the alternative gcc (providing /usr/bin/gcc).

  Selection    Path              Priority   Status
------------------------------------------------------------
  0            /usr/bin/gcc-5     50        auto mode
* 1            /usr/bin/gcc-4.9   49        manual mode
  2            /usr/bin/gcc-5     50        manual mode

but if I good remember, I have checked also gcc-4.8 - with no results...


I'm thinking about recompiling ethminer again under old distro, and trying to make some next tricks
but even if it works, that would be still a shortcut.

[induktor]

Hi Jk_14 good to hear from you!
please re-read my previous post because i edited it, and added a lot of new info.
I almost compile it, by downgrade version of cmake to 3.2.2
now i have some json errors at the end of the compile, but we are close!

about moving the version yeah i did that too and didn't worked, same as you describe it, unresolved symbols.

but we are close to compiling it.

i am thinking maybe this issue is because i had to install the newer libjsonrpccpp-dev
because the older libjson-rpc-cpp-dev cannot be found with apt in 16.04 what do you think?.

can you try what i did so far see if you get same results or go further?

[induktor]

one thing i forgot to explain!

another issue is that OpenCL fails to compile because it cannot find CL/cl.hpp , even if you put it there, it won't see it. (which i did of course!)

so edit CMakeLists.txt
the line that says:

Code:

elseif (BUNDLE STREQUAL "cudaminer")
       set (D_ETHASHCL ON) 

set it to OFF

one less thing to worry about .... no OpenCL allowed here  

--== EDIT ==--

Reading a LOT on the compiling errors about JSON, i found out that it is not a compiling error, the code IS compiling but it is not linking, so i am looking for references to unexisting libraries regarding json, but i am in the dark here, i have no idea where these libraries are.

too bad you flash image get fucked up, can you use the one you give to us? or don't have it anymore?, if that is the case i can make it available, but my upload speed (256Kbps) is a pain and may take like 24 hours to upload.

I have a separate high speed usb flash drive for kopiemtu 3 , from the compiemtu 2 mod #4 which is mining right now in ETH in all my rigs.

on top of that, once i have an operational mining rig, i power it off and do an image file with dd or win32diskimage of every single miner flash drive.
may look excessive but saves a lot of headaches when there is trouble, specially in my area where power outages are very frequent, and flash drives don't like it  

[jk_14]

too bad you flash image get fucked up, can you use the one you give to us? or don't have it anymore?, if that is the case i can make it available, but my upload speed (256Kbps) is a pain and may take like 24 hours to upload.

What kind of problem is with the image?

I made two bug fixing in the fly, not significant, so I didn't change it to e.g. 3.0a,
(just change typo:
"rotate3 " to
"rotate 3" in /etc/logrotate.conf )

but my new image is working...
I'm just re-downloading current image to see if I have any problems.

[PVmining]

it seems Cuda 8.0 is ok for my configuration...

P.S.
I have found some small bug, but easy to remove: after first boot monitor process should be stopped for 1-2 minutes...

Its algo dependent. lbry, skein and blake for example are pretty well on cuda8 with the right miner (tpruvot's release).
But some older algos or older miners could perform bad.

Because of the monitor. I noticed this too. The webinterface was not shown after first start.
After restart the monitor a few timews while switching some algos it worked.


cheers!