Author

Topic: [ANN] ccminer 2.3 - opensource - GPL (tpruvot) - page 147. (Read 500113 times)

legendary
Activity: 2870
Merit: 1091
--- ChainWorks Industries ---

now who on earth has a working eth miner that i can compile Wink ...

hehehe ...

#crysx

I'm on linux too (ubuntu) and had no probs so far. ping me next time you hit on IRC.
but its a waste of time if you try to point your 750tis on ETH - the time of 8mhash per card is over sadly

come and join on #granite ...

im on now - but wont be for too much longer ...

there is this thing called sleep - i dunno what that is lately - but i think i better find out soon Wink ...

tanx pvmining ...

#crysx
sr. member
Activity: 330
Merit: 252

now who on earth has a working eth miner that i can compile Wink ...

hehehe ...

#crysx

I'm on linux too (ubuntu) and had no probs so far. ping me next time you hit on IRC.
but its a waste of time if you try to point your 750tis on ETH - the time of 8mhash per card is over sadly
legendary
Activity: 2870
Merit: 1091
--- ChainWorks Industries ---
Thanks Epsylon3!
your decred miner works very well - a donation is on the way.

I like your elegant way doing real development and innovations!


@myagui. wise words...

@PVmining  Wink

@Epsylon3: Decred donation incoming as soon as my coins unlock by voting (which I learned too late might easily take a month  Grin). Nah, no need to wait that long, whatever I get on Supernova for a few days will be going your way. Decred addy?

job well done in my opinion also ...

now who on earth has a working eth miner that i can compile Wink ...

hehehe ...

#crysx
legendary
Activity: 1154
Merit: 1001
Thanks Epsylon3!
your decred miner works very well - a donation is on the way.

I like your elegant way doing real development and innovations!


@myagui. wise words...

@PVmining  Wink

@Epsylon3: Decred donation incoming as soon as my coins unlock by voting (which I learned too late might easily take a month  Grin). Nah, no need to wait that long, whatever I get on Supernova for a few days will be going your way. Decred addy?
sr. member
Activity: 330
Merit: 252
Thanks Epsylon3!
your decred miner works very well - a donation is on the way.

I like your elegant way doing real development and innovations!


@myagui. wise words...
legendary
Activity: 2870
Merit: 1091
--- ChainWorks Industries ---
i just have released the final 1.7.2 with decred support...

Code:
  Feb. 11th 2015  v1.7.2
                  Decred Algo (longpoll + getwork - no stratum)
                  Blake256 improvements/cleanup

  Jan. 26th 2015  v1.7.1
                  Implement sib algo (X11 + Russian Streebog-512/GOST)
                  Whirlpool speed x2 with the midstate precompute
                  Small bug fixes about device ids mapping (and vendor names)
                  Add Vanilla algo (Blake256 8-rounds - double sha256)

much appreciated epsylon3 ...

#crysx
legendary
Activity: 1484
Merit: 1082
ccminer/cpuminer developer
i just have released the final 1.7.2 with decred support...

Code:
  Feb. 11th 2015  v1.7.2
                  Decred Algo (longpoll + getwork - no stratum)
                  Blake256 improvements/cleanup

  Jan. 26th 2015  v1.7.1
                  Implement sib algo (X11 + Russian Streebog-512/GOST)
                  Whirlpool speed x2 with the midstate precompute
                  Small bug fixes about device ids mapping (and vendor names)
                  Add Vanilla algo (Blake256 8-rounds - double sha256)
legendary
Activity: 1484
Merit: 1082
ccminer/cpuminer developer
sr. member
Activity: 346
Merit: 250
I GOT IT WORKING! Here's what I had to do.

First, install CUDA 7.5.18 from the .run, do not let it install driver 352.39 if you already have 352.63. Also make sure you have g++-4.8 installed.
Navigate to /usr/local/cuda-7.5/include. Comment out lines 113-117 in host_config.h

Code:
113 /*#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 9)*/
114
115 /*#error -- unsupported GNU version! gcc versions later than 4.9 are not supported!*/
116
117 /*#endif*/ /* __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 9) */

In makefile.am use

Code:
nvcc_ARCH  = --compiler-bindir /usr/bin/gcc-4.8 -gencode=arch=compute_50,code=\"sm_50,compute_50\"

Run build.sh. Et Voi-La!

Even though I have the correct versions of gcc and g++ installed, the GNUC version test incorrectly identifies which one is running and screws up the compilation.
legendary
Activity: 1470
Merit: 1114
I already answered
Yes, thank you for your input. However, there is no information in the readme file regarding g++ or Fedora. I build clean every time using build.sh. I have already tried compiling with g++-5.2.1 / CUDA 7.5.18. That's where this all began. I have also seemingly compiled successfully against the repo version of CUDA 6.5 / gcc-4.9 but got an error no nVidia driver installed when I tried to run ccminer. I have one more combination to try before throwing in the towel - Install CUDA 6.5 from the .run file (these always seem to give me the highest success rates) and try compiling with the various gcc versions to see if I can find a combination that works.

My experience is with Fedora so YMMV.

It seems the compiler in ubuntu is not compatible with ccminer. In most cases where a distro
introduces incompatible changes (assuming they are aware) they will also keep the old
compatible version available.

Regarding cuda and driver installation that can be confusing. Quite often when you install cuda
from the repo it will install the repo drivers as well, which may be older. I like to install both
from the nvidia run files to avoid this.

Now, you say you compiled successfully at first but ccminer failed to start because of no driver.
This smells like a nouveau problem. When you installed cuda from the repo you may have
blown away the nvidia proprietary drivers and replaced them with nouveau. I agree you should
try backing out the repo packages then install cuda and the drivers from the run files.
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
I tried to set -i 24 and -i 25 but it results in a crash.. My version works fine with -i 25. (The default is set to 24)

with -i 23 i get 15.82MHASH in the 1.7.1 x86 version.

so my private is only 15% faster with a proper launchconfig.

That's a slightly better comparison, but again, rejects are important to take into account.
If you scroll back a single page on this thread, you'll find a user reporting your (public?) quark doing 11% faster hashrate than tpruvot's, yet it also has 30% rejects doing solo, vs 0% on tpruvot's. So, for those conditions, your faster miner ends up being much slower.

I'm not even putting efficiency on the table, but for some miners, that's an even greater factor of importance. Personally, don't care for a 5% performance boost, at the expense of increasing 10% on power usage.

Side comment: You had my respect and admiration when you started off, making sensible contributions to the open ccminer community. Things have gone downhill ever since the pissing contests began, along with repeated calls for private version donations (with terms that I find unfair to those that actually buy said private versions)...
I do hope you figure out a better way to stay motivated and contributing.

--- apologies for the off topic ---

I need to refork. I have an overflow bug in mine. After the refork i can add etther and decred support
sr. member
Activity: 346
Merit: 250
Still no joy. In all cases where I can get it to compile without puking, when I run ccminer I get
Code:
Unable to query number of CUDA devices! Is an nVidia driver installed?
Same driver I use in all my machines, 352.63.  Undecided
sr. member
Activity: 346
Merit: 250
I already answered
Yes, thank you for your input. However, there is no information in the readme file regarding g++ or Fedora. I build clean every time using build.sh. I have already tried compiling with g++-5.2.1 / CUDA 7.5.18. That's where this all began. I have also seemingly compiled successfully against the repo version of CUDA 6.5 / gcc-4.9 but got an error no nVidia driver installed when I tried to run ccminer. I have one more combination to try before throwing in the towel - Install CUDA 6.5 from the .run file (these always seem to give me the highest success rates) and try compiling with the various gcc versions to see if I can find a combination that works.
legendary
Activity: 1154
Merit: 1001
I tried to set -i 24 and -i 25 but it results in a crash.. My version works fine with -i 25. (The default is set to 24)

with -i 23 i get 15.82MHASH in the 1.7.1 x86 version.

so my private is only 15% faster with a proper launchconfig.

That's a slightly better comparison, but again, rejects are important to take into account.
If you scroll back a single page on this thread, you'll find a user reporting your (public?) quark doing 11% faster hashrate than tpruvot's, yet it also has 30% rejects doing solo, vs 0% on tpruvot's. So, for those conditions, your faster miner ends up being much slower.

I'm not even putting efficiency on the table, but for some miners, that's an even greater factor of importance. Personally, don't care for a 5% performance boost, at the expense of increasing 10% on power usage.

Side comment: You had my respect and admiration when you started off, making sensible contributions to the open ccminer community. Things have gone downhill ever since the pissing contests began, along with repeated calls for private version donations (with terms that I find unfair to those that actually buy said private versions)...
I do hope you figure out a better way to stay motivated and contributing.

--- apologies for the off topic ---
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
I tried to set -i 24 and -i 25 but it results in a crash.. My version works fine with -i 25. (The default is set to 24)

with -i 23 i get 15.82MHASH in the 1.7.1 x86 version.

so my private is only 15% faster with a proper launchconfig.
legendary
Activity: 1154
Merit: 1001
Just tested the latest x86 binary in the quark algo (default intensities)

Gigabyte windforce gtx 970 G1: (1366Mhz core)

ccminer 1.7.1:     14.2MHASH
Sp-mod private #5: 18.2MHASH (+28,17%)

@SP_

How about you try running both by explicitly setting the same value for intensity? Your defaults are known to be much higher than the ones from every other developer, making these stats above rather meaningless.

Also, you've been around long enough that you should know by now, that miner client-side performance is not a reliable metric by itself, some folks have reported your builds to show the faster miner client-side, but have such a high reject rate, that the net result is in effect a much slower miner.

A reliable comparison, is set by running the same parameters (not the defaults), and post the hashrate graph for a reliable pool during a reasonable window on each miner (6 hours or more). A nicehash graph is pretty good for this purpose.

Kindly avoid polluting the threads for other developers with random drops of grossly interpreted hashrate metrics.
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
Just tested the latest x86 binary in the quark algo (default intensities)

Gigabyte windforce gtx 970 G1: (1366Mhz core)

ccminer 1.7.1:     14.2MHASH
Sp-mod private #5: 18.2MHASH (+28,17%)


legendary
Activity: 2870
Merit: 1091
--- ChainWorks Industries ---
epsylon3 ...

are you adding the decred kernel to the miner? ...

if so - when? ... and how would we set the commandline for optimal usage? ...

#crysx
legendary
Activity: 1484
Merit: 1082
ccminer/cpuminer developer
I already answered
sr. member
Activity: 346
Merit: 250
Okay. I was almost ready to report success but not quite. I'm building with gcc-4.9 and I also had to install g++-4.9 separately. The compile looks like it's going to fly, gets all the way through the algos and then this
Code:
#error -- unsupported GNU version! gcc versions later than 4.9 are not supported!
Is there one more option I need to force?
Jump to: