CCminer(SP-MOD) Modded GPU kernels.

[bathrobehero]

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:

[Grout]

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.

[Grout]

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!

[flipclip]

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).

[chrysophylax]

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

[hashbrown9000]

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]

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

[chrysophylax]

sp ...

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

testing now with neoscrypt and obviously donating also - 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

[cooman]

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

[sp_]

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

[chrysophylax]

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

[chrysophylax]

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 ...

#crysx

[sp_]

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

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

thats ok ... ill be testing it all with the donation links once i have them all setup ... ive just been WAY too busy with life at the moment ...

this will all be full time for me soon - so ill have a great deal more time to invest on our farm ...

now - if you mean from djm's optimized ccminer ... i wouldnt know ... sorry sp ...

i never got the chance to mine as the compiled private miner he was distributing was already compiled ( from ubuntu i think - and he was so kind to provide it ) - and the farm wouldnt accept it with the setup that it was in ...

he DID mention to me many a time to upgrade to the latest cuda 6.5 ( i was using and compiling with 6,0 at the time ) BUT i didnt as it was a major undertaking ( as i have now found out ) - so i have no comparison ...

but i do have that private miner in the office server - so i can test it tomorrow if thats what you are after ...

as far as i know - there were also extra parameters like -throughput ... are those paramaters necessary with this spmod fork? ...

these optimizations you are making are getting very good hashrates ...

if you are asking from ccminer itself ... then the answer is very close ...

ccminer itself showed hashrates a little above that ... approx 950KH - 960KH total ...

tanx sp ...

#crysx

[sp_]

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

I have fixed it now@git. The rejects where of the type low difficulty share. X11 also had this problem on some pools

[CapnBDL]

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

I have fixed it now@git. The rejects where of the type low difficulty share. X11 also had this problem on some pools

Ok folks....just a quick question...when will the finished product be ready for D/L?

Thank You for all the hard work you guys are doing.

Happy Mining!!

[antonio8]

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

I have fixed it now@git. The rejects where of the type low difficulty share. X11 also had this problem on some pools

Will compile and run it.

[sp_]

I have fixed the stats for the -g switch. The miner show lower output like we have observed on the pools.
Klaus_t has introduced cudastreams in his fork, I am trying it out to see if it improves the thread perfomance.

[CapnBDL]

Ahhh ha! Got it to do this again. I've been getting disconnected from the pool and would like to know if there is a problem or a workaround to it. I have a single 750Ti card running a basic .bat file. Just the pool with the algo & " -g 2 --cpu-priority 3" switches. I 'sometimes' get this error;

Stratum connection failed: Failed to connect to poolX.com port XXXX: Connection refused
...terminating workio thread


of course the Xs are not really there...but ya get the meaning.
Is this a problem on my end, pool side or a coding issue? This has never been an issue, in the past. Very much appreciated!

Thank You for all the hard work!

Happy mining!!

edit: Oh crap.....running the early D/L of release49, CUDA6.5, Win8.1...sorry, shoulda told ya that up top!

[sp_]

On the fixed version on git we have a boost in quark with -i 22 -g 5. on the 750ti.

Seems to be +250khash per card or more.. The longer the miner is running the bether hashrate on the pool