CCminer(SP-MOD) Modded GPU kernels.

[chrysophylax]

To the linux guys: what version of CUDA are you compiling with?

6.0 ...

we have been playing with 6.5 and it seems to work quite well ...

our miners are now all running 6.5 compiled ccminer and they seem to be stable ...

but the moment a major crash happens - we revert back to the 6.0 compiled miners and they just work ...

#crysx

CUDA 6.5 --

The stuff from SP_ compiles on CUDA 6.5.  I believe  that tpruvot moved to CUDA 7.0, but he takes pains to keep his miners compatible with older nVidia cards.  I don't personally know if CUDA 7.0 can compile programs built for 6.5 properly, although I am curious.  I would install CUDA Toolkit 7.0 on my system if it would work well with the code SP_ produces. CBuchner's CudaMiner was written for CUDA 5.5, tpruvot has a version re-written for 6.5.

I'm looking forward to moving to CUDA 7.0, especially if it produces more hash.       --scryptr

same here - though cuda 7.0 is not compiling anywhere near as compliant as 6.5 is ... for now anyway @scryptr ...

should be interesting to see how this all works a few months down the track when nvidia change the whole playing field with a new chipset once again ...

#crysx

[sp_]

I have added another 9% in skein. Most gain on compute 5.2. Will clean the code abit and submit to github

the  970 is now hashing at 265MHASH.

The total gain from version 46 on the 970 is 38%

[sp_]

Added keccak boost.

Keccak can now be run with intensity 28, before only 21 was possible.

defaults are set to 27 for compute 5.0 and 28 for compute 5,2

[chrysophylax]

Added keccak boost.

Keccak can now be run with intensity 28, before only 21 was possible.

defaults are set to 27 for compute 5.0 and 28 for compute 5,2

most i could run keccak on westhash was -i 24.9 ...

will recompile and see how that goes soon ...

tanx sp ...

#crysx

[sp_]

I have added another 9% in skein. Most gain on compute 5.2. Will clean the code abit and submit to github
the  970 is now hashing at 265MHASH.
The total gain from version 46 on the 970 is 38%

I get some lower rates on the pool. Will investigate.

[sp_]

I have added another 9% in skein. Most gain on compute 5.2. Will clean the code abit and submit to github
the  970 is now hashing at 265MHASH.
The total gain from version 46 on the 970 is 38%

I get some lower rates on the pool. Will investigate.

Submitted some fixes, the hashrate looks good now but get some rejects / duplicate shares.

[sp_]

1.5.47(sp-mod)
@sp-hash sp-hash released this 15 minutes ago

-skein 38% faster (192->265MHASH on the gtx 970)
-Faster Keccak (now support higher intensity values)
-small speedup in quark and scypt

1.5.47(sp-MOD) is available here: (26-april-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.47

The sourcecode is available here:

https://github.com/sp-hash/ccminer

[frazier34567]

I am running 6X750ti EVGA SC (no OC) and for quark when I run -i 22 I get  36744.  When I add -g 2 or any number of threads my CPU goes to 99% and I actually run slower at about 36100-250.   I have tried the release 46 and also compiled my own using commit 757 and both go to 99% cpu.  

I am on windows 8.1 Pro with an i3 processor with 8 GB of RAM.  

Just tried 47 and without an -i or -g parameter I get 3653337120

When I try -i 22 or -i 21 it crashes.

I get nonce errors

[2015-04-26 13:04:34] GPU #5: GeForce GTX 750 Ti, 38196 kH/s
[2015-04-26 13:04:34] GPU #4: result for nonce $769C2CC7 does not validate on CPU!

Then this

Cuda error in func 'cuda_check_cpu_setTarget' at line 28 : unspecified launch failure.

EDIT: Rebooted and now it works up to -i 23

[rednoW]

Forget about -g ... it just doesn't work or has no advantages

[rednoW]

gtx750 oc'ed

#47 vs. Commit 742

skein 83210 vs. 83620
quark 5810 vs 5824 (hashrate fluctuating, maybe incorrect measuring)

conclusion: commit 742 is best for me ...

[ryen123]

Tried the scrypt algo. I get the error seen above. And it just sits there does nothing.

[sp_]

Release 47:

Gtx980 default clock

Skein: 321Mhash

[sp_]

gtx750 oc'ed
#47 vs. Commit 742
skein 83210 vs. 83620
quark 5810 vs 5824 (hashrate fluctuating, maybe incorrect measuring)
conclusion: commit 742 is best for me ...

For skein you should use the latest version. Fixed some bugs in the hash. Checking 32 bits was not enough. Increased to 64 bits.

[sp_]

Tried the scrypt algo. I get the error seen above. And it just sits there does nothing.

Try to add -l 5x24 as a parameter. The autotune seems to have failed

[ryen123]

Ok. Anyway scrypt algo not important.

[scryptr]

RELEASE v47--

Quick Quark stats:

GTX 750ti FTW: 6.32-6.5Mh/s per card, 38.4Mh/s for 6x750ti rig

GTX 960 SSC: 10.1Mh/s per single card in a work computer with low load

GTX 970 FTW+: 14.8Mh/s per card, 29.2Mh/s per 2x970 rig

Error rates are greater.  The last release, v46, could run for hours at 99%+ acceptance rates.  Now, I am getting mid-90% acceptance, early on.

--scryptr

[scryptr]

The as-compiled by sp_ release 45 is a nightmare on my rig, mining Quark

GTX 970
i7 4770
16GB
Win 8.1

Unsteady hashrate with -i 20, and most any other intensity setting crashes.

After a crash, the GPU goes into a low power mode and I have to restart the machine. I suspect that if I used an older version of Afterburner or Precision X that has the K-Boost button, I might be able to wake it up after the crash without restarting .

Either way, it is very buggy and that makes me

download nvidia inspector

Use bat-file with

nvidiaInspector -restartDisplayDriver

Nvidia inspector is an  OverClocking application. Does it have anything to do with recovering from a crash?

DRIVER CRASH--

I am often able to overcome a driver crash with EVGA PrecisionX 16, an overclocking utility.  If you drive the card to hard, and the driver crashes, the card can be reset by turning "Boost Clock" off and on and then re-applying the clocking settings.  The desktop display may also need to be reset, as well.       --scryptr

Thanks scryptr. But what I just installed was TdrDelaySetTo40 that I found in nvminer zip folder. It hasn't crashed yet. But when renting hash and not able to monitor all the time I needed something to restart the miner right away. He is the problem I was having or still maybe have.
 At that time the command window was still open and if I clicked the warning window off the miner would start mining again without any problem.



thx

EDIT: This happened a lot when the stratum failed and at the goto loop. Tested it by disconnecting the ithernet.

:loop
ccminer.exe -r 1 -R 5 -a quark   -o stratum+tcp://us-west01.miningrigrentals.com:3333 -u me -p x  -d 0,1,2
ccminer.exe -r 1 -R 5 -a quark   -o stratum+tcp://us-central01.miningrigrentals.com:3333 -u me  -p x  -d 0,1,2
goto loop

nVidiaInspector --

The program is a GOOD ONE for Windows!  It overclocks and is NOT LIMITED to 4 CARDS!  I am able to access the settings on all 6 of the GTX 750ti SSC cards on the Windows rig, save the settings in a shortcut, and place the shortcut in the Startup Folder on my rig.  Fan speed, clock speed for core and memory, and power settings are all adjustable.  I added 500kh/s to the rig's Lyra2 mining hash rate with the utility (4.5Mh/s now).

@tbearhere --

I also get the Windows warning "ccminer has stopped working" when it cycles through the batch file.  It requires manual intervention and disrupts the functionality of the goto loop.  Apparently, it is a more frequent problem in Windows 8.  I don't have a solution yet, I am still looking.      

 --scryptr

[bathrobehero]

:loop
ccminer.exe -r 1 -R 5 -a quark   -o stratum+tcp://us-west01.miningrigrentals.com:3333 -u me -p x  -d 0,1,2
taskkill -t -f /im ccminer*
timeout -t 1
ccminer.exe -r 1 -R 5 -a quark   -o stratum+tcp://us-central01.miningrigrentals.com:3333 -u me  -p x  -d 0,1,2
goto loop

Should prevent the crash.

[chrysophylax]

:loop
ccminer.exe -r 1 -R 5 -a quark   -o stratum+tcp://us-west01.miningrigrentals.com:3333 -u me -p x  -d 0,1,2
taskkill -t -f /im ccminer*
timeout -t 1
ccminer.exe -r 1 -R 5 -a quark   -o stratum+tcp://us-central01.miningrigrentals.com:3333 -u me  -p x  -d 0,1,2
goto loop

Should prevent the crash.

nice ...

now for a linux one and we will be set ...

...

from memory - there was someone who did actually write a looping script - but cant remember who or when or where ...

#crysx

[scryptr]

:loop
ccminer.exe -r 1 -R 5 -a quark   -o stratum+tcp://us-west01.miningrigrentals.com:3333 -u me -p x  -d 0,1,2
taskkill -t -f /im ccminer*
timeout -t 1
ccminer.exe -r 1 -R 5 -a quark   -o stratum+tcp://us-central01.miningrigrentals.com:3333 -u me  -p x  -d 0,1,2
goto loop

Should prevent the crash.

nice ...

now for a linux one and we will be set ...

...

from memory - there was someone who did actually write a looping script - but cant remember who or when or where ...

#crysx

CBuchner--

CBuchner wrote the looping scrypt for CudaMiner.  He also had a timeout switch that would restart CudaMiner after so many seconds.  Yaamp has a looper that will cycle through the top paying algos.  I revised it to reflect the actual performance of the cards with up-to-date software, and posted a bunch of pages back.       --scryptr