[ANN] ccminer 2.3 - opensource - GPL (tpruvot)

[Epsylon3]

Was unable to reproduce that here... Linux and windows

Are you sure you use my sources... Because ive only seen that in sp version... But not with a wrong

I think it could be a hardware problem which locks cuda

[1714899032]

1714899032

[antimony121]

I compiled mine under compute_30, maybe that make a difference? I am running debug mode to see if i can catch it next time, it seems to behave properly when the host can't be resolved, instead of hanging when it is timed out.

 

Was unable to reproduce that here... Linux and windows

Are you sure you use my sources... Because ive only seen that in sp version... But not with a wrong

I think it could be a hardware problem which locks cuda

[Gimmelfarb]

If you are mining with GPUs, this thread might be of interest for those looking for new GPUs with nice discounts: https://bitcointalk.org/index.php?topic=1207919

[ZeroFossilFuel]

Good day all.
Been trying to assemble your latest (ccminer 1.6.6) for a Kepler (GTX 690), (compute 3.5) .
Getting the following error :
'keccak256_cpu_init' at line 299 : invalid device symbol.
Any advice ?  Or is it simply a Maxwell only scypt?

you need to add compute 3.5 in the projectfile/makefile.

Crispy, did you ever get 1.6.6 working? I'm having the same trouble with a GT640 GDDR5 compute 3.5 card that ran fine with ccminer 1.2/CUDA 5.5/ubuntu 14.04. When I upgraded to Ubuntu 15.04, CUDA went to 6.5 and ccminer 1.2 stopped working.

I uncommented line #544 in makefile to enable compilation of compute 3.5 for my video card.

Code:

./autogen.sh && ./configure && make

 seems to build fine. But executing I get

Code:

[2015-10-15 21:15:51] 1 miner thread started, using 'quark' algorithm.
[2015-10-15 21:15:52] Stratum difficulty set to 0.00390625
Cuda error in func 'quark_blake512_cpu_setBlock_80' at line 245 : invalid device symbol.
~$ 

[myagui]

@ZeroFossilFuel,

Are you refering to some other ccminer fork, or the one on this thread? I'm a bit confused here with that line reference in Makefile.

If I understand correctly, this is the default definition currently on git:
https://github.com/tpruvot/ccminer/blob/windows/Makefile.am#L82

You'd need to comment that line, and uncomment the one for your target architecture, just below (me thinks)... 

[ZeroFossilFuel]

I have both (v1.2) and also the one on this thread (v1.6.6) on Xubuntu 15.04 with Cuda 6.5. The makefile is from 1.6.6. I can not get either to work with my Compute 3.5 video card since upgrading from Xubunru 14.04 to 15.04 and being forced to use Cuda 6.5 vs 5.5. If I could regress to Cuda 5.5 I'm sure my ccminer v1.2 would already be working and I would have abandon all efforts to get 1.6.6 working.

But you're right that I should have been more specific when I stated "I uncommented line #544 in makefile [from 1.6.6] to enable compilation of compute 3.5 for my video card." No joy.

Edit: Now that I'm looking at it, I changed line 544 of makefile, not line 85 of makefile.am. Might that be my problem? Looking back it seems line 544 got commented out again when I recompiled.

[joblo]

@ZeroFossilFuel,

Are you refering to some other ccminer fork, or the one on this thread? I'm a bit confused here with that line reference in Makefile.

If I understand correctly, this is the default definition currently on git:
https://github.com/tpruvot/ccminer/blob/windows/Makefile.am#L82

You'd need to comment that line, and uncomment the one for your target architecture, just below (me thinks)... 

Wrong file, wrong fork. The correct file is Makefile.am before doing anything else. This fork already
has 3.5 enabled, the other one doesn't.

This fork will compile as is for most compute versions. If you want a faster compile and a smaller executable
you can comment out the compute versions you don't need. Just make sure the first one in the list uses the
"=" assignment and all the others use "+=".

[joblo]

I have both (v1.2) and also the one on this thread (v1.6.6) on Xubuntu 15.04 with Cuda 6.5. The makefile is from 1.6.6. I can not get either to work with my Compute 3.5 video card since upgrading from Xubunru 14.04 to 15.04 and being forced to use Cuda 6.5 vs 5.5. If I could regress to Cuda 5.5 I'm sure my ccminer v1.2 would already be working and I would have abandon all efforts to get 1.6.6 working.

But you're right that I should have been more specific when I stated "I uncommented line #544 in makefile [from 1.6.6] to enable compilation of compute 3.5 for my video card." No joy.

Edit: Now that I'm looking at it, I changed line 544 of makefile, not line 85 of makefile.am. Might that be my problem?

See my reply to myagui.

I doubt either the TPruvot fork or the SP_MOD fork will compile with cuda 5.5 so you need cuda 6.5. And either
of these forks support many more algos and produce much higher hashrates than ccminer 1.2, even on older HW.

[ZeroFossilFuel]

Yes, my makefile. was overwritten at compilation because I did not make the change in makefile.am first. I'll try again. Thanks.

[myagui]

Yeah. Go ahead and change the right file and in the right place this time  
It should build and run without issues on your card.

[ZeroFossilFuel]

Still no joy.   

[joblo]

Still no joy.  

The error you're seeing usualy means ccminer can't find a valid cuda GPU. This often occurs
if ccminer is not compiled for the proper compute version but can have other causes such as an
unconfigured card, a faulty card, etc. If the following doesn't work I'm stumped.

The 640 has a monitor attached or is otherwise configured?
It is listed in nvidia-settings?
It is in fact the GDDR5 variation? (GDDR3 is compute 2.1)
You have up to date drivers?
You have cuda 6.5 installed and are configured to use it?
You start with freshly unzipped source?
You edit Makefile.am to ensure code is generated for compute 3.5?
Compiler output confirms 3.5 code is generated?
Have you tried other algos?

Edit: another thought, did you disable the nouveau driver and are use nvidia's proprietary driver?

[chrysophylax]

when i get back - ill test the latest git ...

im curious as to what it can do with some of the main algos i use ... x11 quark neoscrypt ...

#crysx

[Epsylon3]

I recently installed a GTX 460 (SM 2.1) on a machine, and started to do compatibity tests, remains x11 algos to fix but some like lyra have correct speeds (more than a cpu)

I will receive soon a SM 3.0 one to do more tests (GT 740), i only made tests on the 750ti before (compiled as SM 3 only)

[sp_]

Pentablake solomining seem to be broken in 1.7 beta

[Epsylon3]

I dont see any reason... pentablake has nothing special and works on pools

[ZeroFossilFuel]

Still no joy.  

The error you're seeing usualy means ccminer can't find a valid cuda GPU. This often occurs
if ccminer is not compiled for the proper compute version but can have other causes such as an
unconfigured card, a faulty card, etc. If the following doesn't work I'm stumped.

The 640 has a monitor attached or is otherwise configured?
It is listed in nvidia-settings?
It is in fact the GDDR5 variation? (GDDR3 is compute 2.1)
You have up to date drivers?
You have cuda 6.5 installed and are configured to use it?
You start with freshly unzipped source?
You edit Makefile.am to ensure code is generated for compute 3.5?
Compiler output confirms 3.5 code is generated?
Have you tried other algos?

Edit: another thought, did you disable the nouveau driver and are use nvidia's proprietary driver?

I confirm all of these points. It most definitely is GDDR5, Compute 3.5. Otherwise it would have never worked with ccminer 1.2. The only thing on your list I have not tried is other algos.

[Epsylon3]

after the Makefile.am edit, you need to do a proper build, the fastest way is to use ./build.sh

please give me the output of ccminer -n

[ZeroFossilFuel]

after the Makefile.am edit, you need to do a proper build, the fastest way is to use ./build.sh

please give me the output of ccminer -n

GPU #0: SM 3.0 GeForce GT 640

[Epsylon3]

And the algo ? Quark i guess... I will receive a sm3 device soon... Will try to fix it for the 1.7 release