CCminer(SP-MOD) Modded GPU kernels.

[chrysophylax]

i know - just saw it bomba ... tanx for the explanation ... wont work with linux im assuming ... or it will? ...

damn im slow this morning ... :|

will get on irc when i get back in a few hours - gotta run a few errands for now ...

#crysx

#crysx
SP's Release 33 has the ability to set priority and affinity in Linux too (also Windows priority/affinity no longer seems to be hardset in SP's build).  Also per GPU intensity has been added from tpruvot's build.

just compiled and tested ... tanx flipclip ...

seems the fresh algo runs well - but dies with rejects after a little while on yaamp ...

will continue testing when ive completed the move - and reorganized the office ...

tanx ...

#crysx

[1714673012]

1714673012

[1714673012]

1714673012

[chrysophylax]

Nvidia tricked us with GTX 970...
Let's get refund!
https://www.change.org/p/nvidia-refund-for-gtx-970

yeah... I told you, you should have bought the 980gtx 

now regarding the spec, well this is crypto here... people looks more at the price than the spec...
and if they were interested in the spec, they would have gone for the 980gtx...

which is THE reason i went with the gigabyte 750ti oc lp ... no power connector - much more dense than most of the cards out there - and can run 6 cards on one 850 power supply ... with no cables to be tangled everywhere in the mining box ...

o - and they dont overheat - and are much cheaper now ... which is awesome ...



#crysx

[chrysophylax]

Get ready for a 100-600KHASH boost in x11. (750ti) (low difficulty) The 750ti is peaking at 3600KHASH

I am running Echo in parallel with shavite, increasing the AES table cachehits.

With minercontrol software wich switches algos, this is very profitable.

Release 34 is building now.

now THAT is something ive been waiting for

tanx sp ...

#crysx

[flipclip]

i know - just saw it bomba ... tanx for the explanation ... wont work with linux im assuming ... or it will? ...

damn im slow this morning ... :|

will get on irc when i get back in a few hours - gotta run a few errands for now ...

#crysx

#crysx
SP's Release 33 has the ability to set priority and affinity in Linux too (also Windows priority/affinity no longer seems to be hardset in SP's build).  Also per GPU intensity has been added from tpruvot's build.

just compiled and tested ... tanx flipclip ...

seems the fresh algo runs well - but dies with rejects after a little while on yaamp ...

will continue testing when ive completed the move - and reorganized the office ...

tanx ...

#crysx

I believe I was on Fresh at the same time with all the rejects (I think I was getting 30% accept).  I believe it was more yaamp then ccminer though.

[ocminer]

Currently testing on fresh.suprnova.cc - looks good at first.

[Travis9x]

Code:

2015-02-02 19:37:16  2 miner threads started, using 'x15' algorithm.
Cuda error in func 'x11_simd512_cpu_init' at line 641 : invalid texture reference.
*** ccminer 1.5.34-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
	Built with VC++ 2013 and 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. 
19:37:18  Dead YAAMP X15

Seems to happen on all "X" algos.  Both GPUs support Compute 5.0, so have any idea what could be causing this?

EDIT:  Lyra2 algo seems to mine without error, but I'm showing no accepted/rejected shares, and GPU usage sits near-zero…

EDIT 2:  Oddly different error when trying to mine Quark algo -

Code:

2015-02-02 20:31:59  2 miner threads started, using 'quark' algorithm.
Cuda error in func 'cuda_check_cpu_setTarget' at line 28 : invalid device symbol.
*** ccminer 1.5.34-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
	Built with VC++ 2013 and 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. 
20:32:00  Dead YAAMP Quark

[chrysophylax]

Currently testing on fresh.suprnova.cc - looks good at first.

that is your pool is it not ocminer? ...

#crysx

[sp_]

Code:

2015-02-02 19:37:16  2 miner threads started, using 'x15' algorithm.
Cuda error in func 'x11_simd512_cpu_init' at line 641 : invalid texture reference.
*** ccminer 1.5.34-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
	Built with VC++ 2013 and 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. 
19:37:18  Dead YAAMP X15

Seems to happen on all "X" algos.  Both GPUs support Compute 5.0, so have any idea what could be causing this?

EDIT:  Lyra2 algo seems to mine without error, but I'm showing no accepted/rejected shares, and GPU usage sits near-zero…

EDIT 2:  Oddly different error when trying to mine Quark algo -

Code:

2015-02-02 20:31:59  2 miner threads started, using 'quark' algorithm.
Cuda error in func 'cuda_check_cpu_setTarget' at line 28 : invalid device symbol.
*** ccminer 1.5.34-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
	Built with VC++ 2013 and 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. 
20:32:00  Dead YAAMP Quark

you get this error message if one of your cards doesn't support compute 5.0 or higher. Try to run ccminer with -d 0, and then -d 1 and then -d 2 etc. to find wich device that is failing. You can recompile with compute 3.0 / sm 3.0 it it will probobly work

[sp_]

I believe I was on Fresh at the same time with all the rejects (I think I was getting 30% accept).  I believe it was more yaamp then ccminer though.

Fresh is working fine on Yaamp now. I get 3400-3500KHASH on the 750ti.

[sp_]

pls sp, can you help im having trouble with driver crashing with only 970 gtx cards with all of your realases mining mjollnir algo on diferent systems and different os i have TDP at 52% and gpu usage at 91-92%, i have 1200w gold psu per 3 card setups and all cards are stock - non oced and never get hotter than 50 C, i tried all drivers that are out and same cards are not crashing when mining 11 algo or with spreadminer. They are at 61% TDP when mining spread and that is maybe why they are chrashing while mining mjollnir at only 52% maybe they need more power idk can you help, Thanks

Seems to work in benchmark mode. Try to mine a while with only one card. -d 0.

then add another card. -d 0,1

It could be a riser problem.

[Travis9x]

Code:

2015-02-02 19:37:16  2 miner threads started, using 'x15' algorithm.
Cuda error in func 'x11_simd512_cpu_init' at line 641 : invalid texture reference.
*** ccminer 1.5.34-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
	Built with VC++ 2013 and 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. 
19:37:18  Dead YAAMP X15

Seems to happen on all "X" algos.  Both GPUs support Compute 5.0, so have any idea what could be causing this?

EDIT:  Lyra2 algo seems to mine without error, but I'm showing no accepted/rejected shares, and GPU usage sits near-zero…

EDIT 2:  Oddly different error when trying to mine Quark algo -

Code:

2015-02-02 20:31:59  2 miner threads started, using 'quark' algorithm.
Cuda error in func 'cuda_check_cpu_setTarget' at line 28 : invalid device symbol.
*** ccminer 1.5.34-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
	Built with VC++ 2013 and 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. 
20:32:00  Dead YAAMP Quark

you get this error message if one of your cards doesn't support compute 5.0 or higher. Try to run ccminer with -d 0, and then -d 1 and then -d 2 etc. to find wich device that is failing. You can recompile with compute 3.0 / sm 3.0 it it will probobly work

I have a 780Ti and a 760, and running GPU-Z reports they both support Compute 5.0.  I just updated the drivers, and using -d to separate and test the cards one at a time shows the same error with both cards.  Any recommendations?  As far as I can tell, this is the only version of ccminer I can find which supports the Lyra2 algo (and all your other algo optimizations are also very helpful), so I'd really like to get this to work.  Thanks! 

[Bombadil]

I have a 780Ti and a 760, and running GPU-Z reports they both support Compute 5.0.  I just updated the drivers, and using -d to separate and test the cards one at a time shows the same error with both cards.  Any recommendations?  As far as I can tell, this is the only version of ccminer I can find which supports the Lyra2 algo (and all your other algo optimizations are also very helpful), so I'd really like to get this to work.  Thanks!  

Wouldn't be the first time GPU-Z is wrong on anything
Check your GPU here, under "Cuda-Enabled Geforce Products": https://developer.nvidia.com/cuda-gpus
But I can already tell you a 780ti is compute 3.5 and a 760 is 3.0

[sp_]

To make it work, change from compute_50 sm50 to compute_30  / 35 and recompile. I tested some time ago on a gtx 650ti. (compute3.0) Got some hw errors in x15. Not sure if the latest version will work.

you need cuda 6.5 and visual studio 2013.

For compute 3.0 and 3.5 I reccomend this fork:

https://bitcointalk.org/index.php?topic=770064.0

[rednoW]

K4000 is using Compute 3.0. Should work If you build yourself. My binaries are always build with compute 5.0 and 5.2.

just changing settings to compute_30,sm_30 in CUDA C/C++ -> Device -> Code Generation in VS2013 project preferences
leads to results not approved by CPU.
Not that i need it, just tried ))).

As well as trying to compile with CUDA 7.0

Any step from default config leads to not working code ...

[StuffOfInterest]

Interesting to see bitcoin (sha256) support added.  No chance of profit but fun to have most of the algos supported in one miner.  Of note, this doesn't seem to be a very optimal implementation.  With cgminer (opencl) I was getting about 405,000 kh/s on my 2 x 750 Ti setup.  With release 34 ccminer I'm down to about 345,000 kh/s.  Still, it sits at the very bottom of the profitability list with either hash rate so I'm not too concerned.  It burns many more times in power than it earns.

[sp_]

Interesting to see bitcoin (sha256) support added.  No chance of profit but fun to have most of the algos supported in one miner.  Of note, this doesn't seem to be a very optimal implementation.  With cgminer (opencl) I was getting about 405,000 kh/s on my 2 x 750 Ti setup.  With release 34 ccminer I'm down to about 345,000 kh/s.  Still, it sits at the very bottom of the profitability list with either hash rate so I'm not too concerned.  It burns many more times in power than it earns.

The code was added by klaus-t. Probobly a pure C port without any cuda.
The code could be optimized, but it doesn't matter.  Even if I manage to push the speed to 1GHASH---> not profitable.

But perhaps I will try anyway...

I will add some more speedup in x13,x14,x15 today.  Similar to the fresh/x11 optimalizations. Divide the stream into two streams and run in paralell.

Nobody mining x11 anymore? what rates are you getting? (ppol vs miner) (970 and 980 is untested so please test)

Use a low difficulty

on nicehash.com you can add the difficulty in the password like this: -p=d=0.02

[sp_]

K4000 is using Compute 3.0. Should work If you build yourself. My binaries are always build with compute 5.0 and 5.2.

just changing settings to compute_30,sm_30 in CUDA C/C++ -> Device -> Code Generation in VS2013 project preferences
leads to results not approved by CPU.
Not that i need it, just tried ))).
As well as trying to compile with CUDA 7.0
Any step from default config leads to not working code ...

yeah. It might work on compute 3.5 though..

Use this version (1.5.2-git)

https://bitcointalk.org/index.php?topic=770064.0

it works.

[rednoW]

yeah. It might work on compute 3.5 though..

Use this version (1.5.2-git)

https://bitcointalk.org/index.php?topic=770064.0

it works.

yes, it works. Good in "fresh".
Strange but for "quark" on compute 3.0 original ccminer 1.2 is the best, just for fun


Tried to build your release 34 in x64 mode. Small performance drop in all algos.
Test comp: Athlon64X2 3ghz, gtx 750 (1470gpu/1550mem), win7x64

rates on your original release34 build are
fresh: ~3450kh  
quark: ~5550kh (on release 31 is ~5600kh) up to 61% of modded TDP 65w (cool)

release 34 seems to work better with yaamp then release 31

[bathrobehero]

K4000 is using Compute 3.0. Should work If you build yourself. My binaries are always build with compute 5.0 and 5.2.

just changing settings to compute_30,sm_30 in CUDA C/C++ -> Device -> Code Generation in VS2013 project preferences
leads to results not approved by CPU.
Not that i need it, just tried ))).

As well as trying to compile with CUDA 7.0

Any step from default config leads to not working code ...

I can confirm that compiling for 3.0 doesn't work (CPU validation errors) but 3.5 is fine.

For anyone wondering about compute versions:

3.0   GeForce GTX 770, GeForce GTX 760, GeForce GT 740, GeForce GTX 690, GeForce GTX 680, GeForce GTX 670, GeForce GTX 660 Ti, GeForce GTX 660, GeForce GTX 650 Ti BOOST, GeForce GTX 650 Ti, GeForce GTX 650, GeForce GTX 780M, GeForce GTX 770M, GeForce GTX 765M, GeForce GTX 760M, GeForce GTX 680MX, GeForce GTX 680M, GeForce GTX 675MX, GeForce GTX 670MX, GeForce GTX 660M, GeForce GT 750M, GeForce GT 650M, GeForce GT 745M, GeForce GT 645M, GeForce GT 740M, GeForce GT 730M, GeForce GT 640M, GeForce GT 640M LE, GeForce GT 735M, GeForce GT 730M, Quadro K5000, Quadro K4200, Quadro K4000, Quadro K2000, Quadro K2000D, Quadro K600, Quadro K420, Quadro K500M, Quadro K510M, Quadro K610M, Quadro K1000M, Quadro K2000M, Quadro K1100M, Quadro K2100M, Quadro K3000M, Quadro K3100M, Quadro K4000M, Quadro K5000M, Quadro K4100M, Quadro K5100M, Tesla K10

3.5   GeForce GTX TITAN Z, GeForce GTX TITAN Black, GeForce GTX TITAN, GeForce GTX 780 Ti, GeForce GTX 780, GeForce GT 640 (GDDR5), GeForce GT 630 v2, Quadro K6000, Quadro K5200, Tesla K40, Tesla K20x, Tesla K20

3.7   Tesla K80

5.0   GeForce GTX 750 Ti, GeForce GTX 750, GeForce GTX 860M, GeForce GTX 850M, GeForce 845M, GeForce 840M, GeForce 830M, Quadro K2200, Quadro K620, Quadro K620M

5.2   GeForce GTX 980, GeForce GTX 970, GeForce GTX 960, GeForce GTX 980M, GeForce GTX 970M, GeForce GTX 965M

[sp_]

yeah. It might work on compute 3.5 though..
Use this version (1.5.2-git)
https://bitcointalk.org/index.php?topic=770064.0
it works.

yes, it works. Good in "fresh".
Strange but for "quark" on compute 3.0 original ccminer 1.2 is the best, just for fun
Tried to build your release 34 in x64 mode. Small performance drop in all algos.
Test comp: Athlon64X2 3ghz, gtx 750 (1470gpu/1550mem), win7x64
rates on your original release34 build are
fresh: ~3450kh  
quark: ~5550kh (on release 31 is ~5600kh) up to 61% of modded TDP 65w (cool)
release 34 seems to work better with yaamp then release 31

Release 31 uses more CPU, but if your cpu is good, more hash.
Compute 3.0 uses less registers and have different assembly instructions, not strange that it goes slower than the old version. All my modifications are tested on the Maxwell chips only. Compute 5.0 +
x64 has always been slower on windows. Mine is buildt with x86.
What rates are you getting in x11?