CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1104.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: chrysophylax on May 06, 2015, 11:42:58 PM

sp ...
finally got it all sorted with compile and setup of the new os / install ...
testing now with neoscrypt and obviously donating also Wink

- http://yaamp.com/?address=1CTiNJyoUmbdMRACtteRWXhGqtSETYd6Vd ...
-i 15.8 ( 58880 cuda threads - max it will go on this system ) ... -a neoscrypt -i 15.8
fedora 20 x64 - 6 x gigabyte 750ti oc lp cards - cuda 6.5 - latest git clone spmod ...
will let it run for an hour or so - and see how it fares ...
any settings that i should know about that would optimize it further? ...
tanx ...
#crysx

Tnx for using my btc account for testing.

In the short test period 6x750ti hashed neoscrypt from

814-1231Khash

The average rate in the end was 964 KHASH. (161 Khash per card.) Is this close to the output you got from ccminer?

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

hi all ...

found what the issue was ...

only the nvidia drivers were required to be removed from the original install BEFORE the negativo17 repo was to be installed - and not all the cuda 6.5 files as well ...

they apparently work together - as both repos can be installed alongside the other and are compatible ...

now testing the install and the performance of the miner in the newest installation ...

what a learning curve for me ... sheesh! ...

sorry to put you guys all through this adventure of mine also - and flooding the forum sp ... but this just HAS to be helpful in anyway to some one ...

thats my excuse and im sticking to it Wink

...

#crysx

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

sp - am i doing something wrong here? ...

the last git clone compiled fine ...

-------

g++ -DHAVE_CONFIG_H -I. -fopenmp -pthread -fno-strict-aliasing -I/usr/local/cuda/include -DUSE_WRAPNVML -DSCRYPT_KECCAK512 -DSCRYPT_CHACHA -DSCRYPT_CHOOSE_COMPILETIME -O3 -march=native -D_REENTRANT -falign-functions=16 -falign-jumps=16 -falign-labels=16 -MT ccminer-sysinfos.o -MD -MP -MF .deps/ccminer-sysinfos.Tpo -c -o ccminer-sysinfos.o `test -f 'sysinfos.cpp' || echo './'`sysinfos.cpp
ccminer.cpp:12:30: fatal error: cuda_runtime_api.h: No such file or directory
#include "cuda_runtime_api.h"
^
compilation terminated.
make[2]: *** [ccminer-ccminer.o] Error 1
make[2]: *** Waiting for unfinished jobs....
nvml.cpp:27:26: fatal error: cuda_runtime.h: No such file or directory
#include "cuda_runtime.h"
^
compilation terminated.
make[2]: *** [ccminer-nvml.o] Error 1
mv -f .deps/ccminer-sysinfos.Tpo .deps/ccminer-sysinfos.Po
mv -f .deps/ccminer-stats.Tpo .deps/ccminer-stats.Po
make[2]: Leaving directory `/mnt/compile/ccminer-all/sp-hash/ccminer'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/mnt/compile/ccminer-all/sp-hash/ccminer'
make: *** [all] Error 2

-------

tanx ...

#crysx

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Yes I can. Scrypt -n doesn't seem to work in the latest mod. Benchmarkmode reports no nonces.

cooman

newbie

Activity: 8

Merit: 0

Hi SP_,Could you optimized the scrypt-jane algorithm? THX! Grin

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

sp ...

finally got it all sorted with compile and setup of the new os / install ...

testing now with neoscrypt and obviously donating also Wink

- http://yaamp.com/?address=1CTiNJyoUmbdMRACtteRWXhGqtSETYd6Vd ...

-i 15.8 ( 58880 cuda threads - max it will go on this system ) ... -a neoscrypt -i 15.8

fedora 20 x64 - 6 x gigabyte 750ti oc lp cards - cuda 6.5 - latest git clone spmod ...

will let it run for an hour or so - and see how it fares ...

any settings that i should know about that would optimize it further? ...

tanx ...

#crysx

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

Quote from: hashbrown9000 on May 06, 2015, 10:09:29 PM

hmm, running ubuntu 14.04 i never had a no cuda devices installed error. as long as the path is exported each time before the miner is run, the miner works on any user including root.

maybe cuda installs the path environment automatically in ubuntu ...

i know it doesnt in fedora - it needs to be explicitly set ( like you have mentioned ) in either the script or the environment settings ( .bashrc ) ...

#crysx

hashbrown9000

sr. member

Activity: 427

Merit: 250

hmm, running ubuntu 14.04 i never had a no cuda devices installed error. as long as the path is exported each time before the miner is run, the miner works on any user including root.

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

this happened with the fedora 20 x64 install last night ( actually 0300 in the morning ) ...

exactly the same situation with exactly the same results ...

i went around it a little differently - but same result ...

we now have a fedora 20 x64 machine running the latest ccminer-spmod with cuda 6.5 ...

after almost half a day researching and testing - the light at the end of the tunnel was purely a driver and installation issue ...

using the very same drivers FROM nvidia themselves that caused most of the problems - i managed to install various components separately - only to find a dev had already done the leg work for this installation - AND created rpm's and a repo for the installs ...

negativo17 - thats what eventually saved the ( installation ) day ... http://negativo17.org/nvidia-driver/ ...

then that issue with cuda devices not being found ... that was resolved ( for some reason ) with the installation of the nvidia-driver-NVML ... only to find that there is now NVML GPU monitoring enabled ...

that is a first for a fedora installation ...

once all of the base was installed - the compile was done successfully and now the system runs really well ... though the intensity had to be dropped to 20.5 ( originally 20.6 ) with x11 and the other algos have yet to be tested ...

once all the errands are run today - the rest of the testing will continue ...

hope this helps those of you in the same boat as we were ...

#crysx

flipclip

member

Activity: 111

Merit: 10

Quote from: Grout on May 06, 2015, 05:59:44 PM

Thanks to soka on #ccminer, here is the solution for Ubuntu 15.04 users : you have to create the file AND load the nvidia_uvm module at boottime (which for some reason is not loaded by default on Ubuntu 15.04). Apparently, ccminer can load the module by itself, but it only works as root, obviously. So just edit these two files and you're golden:

/etc/modules

Code:

nvidia_uvm

/etc/rc.local

Code:

mknod -m 666 /dev/nvidia-uvm c 249 0

Cheers!

Thanks for this info. Works on Ubuntu 14.04 too. Had always bugged me, but I'd always forget to look into it (don't restart all that much).

Grout

full member

Activity: 139

Merit: 100

Thanks to soka on #ccminer, here is the solution for Ubuntu 15.04 users : you have to create the file AND load the nvidia_uvm module at boottime (which for some reason is not loaded by default on Ubuntu 15.04). Apparently, ccminer can load the module by itself, but it only works as root, obviously. So just edit these two files and you're golden:

/etc/modules

Code:

nvidia_uvm

/etc/rc.local

Code:

mknod -m 666 /dev/nvidia-uvm c 249 0

Cheers!

Grout

full member

Activity: 139

Merit: 100

Hi,

I have a problem on Ubuntu 15.04, that happens with any version of ccminer (SP, tpruvot, djm34).

The compilation went fine, but on the first run, and after each reboot, I get:

Code:

julien@miner1:~/ccminer-SP$ ./ccminer --benchmark -a x11
*** ccminer 1.5.49-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
   Built with the nVidia CUDA SDK 6.5

  Based on pooler cpuminer 2.3.2 and the tpruvot@github fork
   CUDA support by Christian Buchner, Christian H. and DJM34
  Includes optimizations implemented by sp , klaust, tpruvot and tsiv.

[2015-05-06 23:54:30] Unable to query number of CUDA devices! Is an nVidia driver installed?

Then, if I start the miner as root, it runs fine:

Code:

julien@miner1:~/ccminer-SP$ sudo ./ccminer --benchmark -a x11
[sudo] password for julien:
*** ccminer 1.5.49-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
   Built with the nVidia CUDA SDK 6.5

  Based on pooler cpuminer 2.3.2 and the tpruvot@github fork
   CUDA support by Christian Buchner, Christian H. and DJM34
  Includes optimizations implemented by sp , klaust, tpruvot and tsiv.

[2015-05-06 23:54:52] NVML GPU monitoring enabled.
[2015-05-06 23:54:52] 2 miner threads started, using 'x11' algorithm.
[2015-05-06 23:54:52] Thread 0 priority 0 (set to 18)
[2015-05-06 23:54:52] Thread 1 priority 0 (set to 18)
[2015-05-06 23:54:52] Binding thread 1 to cpu 1 (mask 2)
1
[2015-05-06 23:54:52] Binding thread 0 to cpu 0 (mask 1)
0
[2015-05-06 23:54:57] GPU #1: GeForce GTX 750 Ti, 2841
[2015-05-06 23:54:57] Total: 5242 kH/s

After that, I can start the miner as a normal user:

Code:

julien@miner1:~/ccminer-SP$ ./ccminer --benchmark -a x11
*** ccminer 1.5.49-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
   Built with the nVidia CUDA SDK 6.5

  Based on pooler cpuminer 2.3.2 and the tpruvot@github fork
   CUDA support by Christian Buchner, Christian H. and DJM34
  Includes optimizations implemented by sp , klaust, tpruvot and tsiv.

[2015-05-06 23:55:23] NVML GPU monitoring enabled.
[2015-05-06 23:55:23] Thread 0 priority 0 (set to 18)
[2015-05-06 23:55:23] Binding thread 0 to cpu 0 (mask 1)
0
[2015-05-06 23:55:23] Thread 1 priority 0 (set to 18)
[2015-05-06 23:55:23] Binding thread 1 to cpu 1 (mask 2)
1
[2015-05-06 23:55:23] 2 miner threads started, using 'x11' algorithm.

My guess is, it is a permission issue. It needs to create a file that only root can create. But since the verbose mode doesn't work on the first run, I can't tell where the problem comes from.
I noticed a file that appeared after the "root run" (/dev/nvidia-uvm), but creating it by hand with 'mknod -m 666 /dev/nvidia-uvm c 249 0' is not enough. Must be missing something else...

If it is distribution-specific and nobody has any idea how to fix that, I might downgrade to 14.04.

bathrobehero

legendary

Activity: 2002

Merit: 1051

ICO? Not even once.

Quote from: ?? on ??

Ouch! yaamp hitting quark with 8% fee.

This is not going to be popular but I have to say I like that. I think that serves people right for being lazy. I mean that fee at least incentivizes miners to do their research about what to mine instead of pretty much ruinning the prices of coins they don't even know existed. There are 4 quark coins being mined by yaamp and the combined daily volume of those coins are below 2 BTC and they are all going down:

Slava_K

hero member

Activity: 677

Merit: 500

Cant use sp-mod ccminer on lga775 intel platform. Only old djm34 mod...
p.s. win7 x64, clean install, latest drivers, vc 2013 runtimes installed.

antonio8

legendary

Activity: 1400

Merit: 1000

@ chrysophylax

No -g parameter. It does not work for me as it just crashes my driver, Windows 7 64 bit. It does the same on Windows 10 64 bit. Just running regular bat. Not even -i setting.

@ sp_

That build does have a slight speed increase and my hash rat is a little higher after about 10 minutes of running but still only 88% so far.

Will see if it gets better.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: antonio8 on May 06, 2015, 06:31:42 AM

I get low accepted % on lyra2. Around 84% on release 49.

Try this build, or build yourself.

http://cryptomining-blog.com/4718-updated-windows-binary-of-the-ccminer-1-5-49-git-fork-by-sp-for-maxwell/

There has been some fixes/improvements in lyra2

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

Quote from: antonio8 on May 06, 2015, 06:31:42 AM

I get low accepted % on lyra2. Around 84% on release 49.

are you using ccminer-spmod with the -g option? ...

if so - drop it and go back to the 'standard' parameters and see how that works mate ...

#crysx

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

Quote from: djm34 on May 06, 2015, 05:11:17 AM

Quote from: chrysophylax on May 05, 2015, 11:49:59 PM

Quote from: djm34 on May 05, 2015, 09:58:38 PM

Quote from: chrysophylax on May 05, 2015, 09:22:56 PM

hi all ...

been busy trying to figure out what it is that the issue maybe with the compilation lock up and freezes since the addition of neoscrypt in ccminer-sp ...

with the help of a fantastic ( but for the moment hidden - you know who you are mate ) watcher of this thread - it looks like we have narrowed down the issue to versions of cuda ( and some associated files ) in fedora itself ...

if this is the case - and it really seems like it is - then a full upgrade of the farm needs to be taken into account ... something we were not looking forward to ... a full weeks worth of work ...

djm34 pointed a similar issue out quite a while ago when he still had the miner in closed distribution ( cuda version 6.0 - cuda 6.5 ) - whereby his neoscrypt miner refused to compile in the same environment ...

yep, I told you so repeatedly Grin

must admit, I never understood your insistence at that level for keeping a deprecated package, cuda is a dev package for nvidia (not related to any specific drivers, neither to the os)... anyway...
I guess many have to tell you the same thing for you to listen... Grin

no - not at all djm ...

its the working model we have for the farm ... we are not just a single user with a computer or two that just mine occasionally djm ... we have a certain protocol we follow in order to make things digestible for easy implemenatation for the farm that are under the guidelines we have set for the farm ...

even so - ccminer-spmod is STILL not working with cuda 6.5 - but it compiles now ...

its the distro we have chosen ( fedora IS bleeding edge - so its expected to have bugs and incompatabilities ) ... and the reason we have chosen THIS distro is due to its ease of use as a desktop environment ...

our farm is NOT just for farming - its for ease of upgrade ... fedora is the 'better' package for that ...

if the farm is to be pulled apart for instance - and sold off - the machines require no more than a few packages removed and another few installed for it to be built into a computer system that can be sold off for the basic user ... a matter of 20 minutes work - unlike a completely full install with additions of hdd and a few other components .. these machines are COMPLETE in every way as a desktop system except for the software ( and 5 or 6 cards per machine ) ... our goal is for adaptability for the farm and unfortunately - that is fraught with issues ...

the new farm ( upgraded farm we have been working on for the last few months ) is SPECIFICALLY a farm - even though it has a similar goal for updatability - it is designed only to be upgraded as a farm and nothing else ... so the apps and versions and way its built ( from custom designed frames to 'standard' software and apps rolled out across the entire farm ) is standard right down to the components used ... ALL machines will be EXACTLY the same ...

so - one compile - one miner - one installation ...

we are trying to save a lot of work upgrading THIS current farm ( especially me ) - so deprecated or not - it is what we were aiming at using different components ( different cpus - ram - motherboards psu ) ... this farm is what we have until the other farm is ready ... once the new farm is ready - it will be very easy to replicate every machine - both hardware and software ... and it will all have the latest software - so we can keep up Wink

...

so you are correct in that we need to sustain updatability - but even with the successful compile - the miner is still not working ...

we cannot be the ONLY ones with this issue - considering fedora is one of the largest distributions for desktops in the world ... maybe those with these issues also are just not speaking up

...

when we get the miner running properly under fedora ( 20 or 21 ) - then the whole process can be replicated easily ...

until then - v47 is the version that is running currently with v48 on a couple of the other systems under fedora 19 x64 ...

cuda 7.0 repo is about to be released under fedora 21 x64 ... that is the one im more interested in Wink

... as apparently it has backward compatibility to the previous versions ...

i personally want an automated system for farmers ... which will mean that even those that just want to run only a couple of systems - will be able to duplicate the method and have miners / workers AND desktops / development machines with a single install ...

i would luv to be able to have a management console that could be rolled out ( in a single .rpm ) across an entire system for stats and maintenance purposes ... but that is a bit down the track ...

btw - we cannot use 'other' systems like kompiemtu and the like unless we can convert these machines back to desktop systems ... ideas are always welcome though Wink

...
#crysx

actually kopiemtu and alike work from a usb stick or very small drive, and wouldn't alter your system at all, actually it would be even better for you, since you could keep your system and have a dedicated os for mining you could run or not.

agreed ...

but with so many systems being built and running them on a system like that ( one which we have very little experience with ) the learning curve is a steep and long one ...

you do however - make a very strong point about it being a separate system to run - without touching the base system ...

i personally have looked into it - and find it fascinating in that it is already set as an 'as is' miner ...

really - the only issue that i have found with it - is as soon as you start changing bits in the software - you 'can' run into issues with driver compatibility and so forth ...

we have a copy of it here - so when the testing and building and reconstruction of the farm is done - i will definitely look into it further in much more depth ...

tanx djm ...

#crysx

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

Quote from: fenomenhaa on May 06, 2015, 04:35:39 AM

Quote from: chrysophylax on May 06, 2015, 04:20:30 AM

Quote from: fenomenhaa on May 06, 2015, 04:11:55 AM

Quote from: sp_ on May 06, 2015, 01:33:10 AM

with 6-7 970 cards at 200w you need 1500Watt of power for your rig. Cheaper to spread the cards, 2-3 970 and fill the rest of the pci slots with 960 or 750ti cards.
Then you can use normal cheaper gaming PSU's

...

http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-970/specifications:

The GeForce GTX 980 has a TDP rating of just 165 Watts, and the GTX 970 145Watt.

Maybe better to spread the psu's

))

that was one of the main considerations with building our workers ...

we have cut psu usage down from high powered 1500W psu - to standard 850W thermaltake due to the use of the gigabyte 750ti oc lp cards ...

the new systems will be built with 1000W corsair rm1000 psu ( as they are a low profile managed psu also ) - to power all 6 x 750ti oc lp cards and the rest of the machine with plenty of ( power ) room to spare ...

#crysx

Corsair Rm1000 W psu is too much for 6 x750Ti rig.RM1000W is Chicony built good quality psu .Easliy configurable with 12 750Ti

)) .Usualy regular good quality psu's have best efficients between %50-80 load.According to wattmeter 5x750Ti (Bios moded Tdp incerase to 65W +o/c 150core 100mem vomoded to 1.2v)390W from the wall(Quark) other algos consumption is less then quark With a %85 efficeny psu so its :330W in reality. My advice is 600-750w (strongly recommend Seasonic or enermax builds) Psu strong and very effiecient option with 6x750Ti rigs.

thats the whole point ... the rigs ( custom designed frames and systems ) will have exactly 12 x gigabyte 750ti oc lp units running from it ...

building it this week ... as it has passed all the testing that we have thrown at it - and is not that expensive for a good quality unit ...

no modding on ANY of the cards ... we like to stick with stock cards so there are no hidden surprises ... the moment you start modding and playing with anything else bar the clocks - you are putting yourself in a situation of uncharted territory - and may cause a great deal of loss of time and effort ... valuable time that could have gone into building a better system - and much more of them Wink

...

for those that want to mod and play with firmware - no issues ... we cannot afford that luxury of the untested grounds ... yet Wink

...

currently - its a testing phase for fedora 20 x64 and all thats involved with running these systems efficiently and building a practical working system as a base ... THEN when the whole farm is done - start to work on mods and such ...

high efficiency psu like the rm1000 will be a long term investment for the miners ... less downtime and much less maintenance ...

we have been running thermaltake 875W 850W as well as 750W and a few other psu's ( size and brands ) only to find intermittent issues with these lower end psu's ... still good for the short term - but not very efficient ...

each rig will be a hashrate junkie - with potential to be upgraded to MUCH bigger systems if need be ...

this current farm we have has taught us a great deal ... Wink

...

#crysx

antonio8

legendary

Activity: 1400

Merit: 1000

I get low accepted % on lyra2. Around 84% on release 49.

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1104. (Read 2347601 times)