CCminer(SP-MOD) Modded GPU kernels.

[pallas]

- neoscrypt +10khash on the 750ti (6%)
- qubit +50khash (compute 5.2)
- faster x11, quark (compute 5.2) (quark +300khash on the gtx 980) (1.7 %)
- yesscrypt has been removed.(faster to mine with cpu)
1.5.58(sp-MOD) is available here: (09-08-2015)
https://github.com/sp-hash/ccminer/releases/
The sourcecode is available here:
https://github.com/sp-hash/ccminer

Thanks SP_!
Last commit on neoscrypt is from 15 days ago (just before the previous release), are you sure you committed it all?

I had one commit rigth after the release 57 (10khash on the 750ti) as well.

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

uh sorry!
"neoscrypt +10KHASH gtx 970"
so it's on 970, not 750ti ;-)

[sp_]

yeah. I corrected the release notes.

quark is faster on the 750ti as well.  (from release 57 to 58 on One of my test rigs the gain is 70khash per card.)

X11 needs to be finetuned.

The current default intensity for the gtx750ti is
-i 5 and for  compute 5.2  -i 22

Please help me find the best intensity settings people

[facanferff]

Hi, I'm having problems running it with scrypt:

>ccminer -o stratum+tcp://ca.simplemulti.com:3433 -u XXXXXXXXX -p YYYYYYYY -a scrypt
*** ccminer 1.5.58-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.

[2015-08-09 16:18:47] Starting Stratum on stratum+tcp://ca.simplemulti.com:3433
[2015-08-09 16:18:47] NVAPI GPU monitoring enabled.
[2015-08-09 16:18:47] 1 miner thread started, using 'scrypt' algorithm.
[2015-08-09 16:18:47] Binding thread 0 to cpu 0 (mask 1)
[2015-08-09 16:18:47] Stratum difficulty set to 256
[2015-08-09 16:18:47] GPU #0: GeForce GTX 850M with SM 5.0
[2015-08-09 16:18:47] GPU #0: interactive: 0, tex-cache: 0, single-alloc: 0
[2015-08-09 16:18:47] GPU #0: 32 hashes / 4.0 MB per warp.
[2015-08-09 16:18:48] GPU #0: Performing auto-tuning, please wait 2 minutes...
[2015-08-09 16:18:48] GPU #0: Err 2: out of memory (salsa_kernel.cu:516)
[2015-08-09 16:18:48] GPU #0: Err 2: out of memory (salsa_kernel.cu:517)
[2015-08-09 16:18:48] GPU #0: Err 11: invalid argument (salsa_kernel.cu:522)
[2015-08-09 16:18:48] GPU #0: maximum total warps (BxW): 416

How can I bypass this problem?
Thanks

EDIT:
I tried the latest cudaminer version from github and it seems to work with it:

           *** CudaMiner for nVidia GPUs by Christian Buchner ***
                     This is version 2014-02-28 (beta)
        based on pooler-cpuminer 2.3.2 (c) 2010 Jeff Garzik, 2012 pooler
            Cuda additions Copyright 2013,2014 Christian Buchner
          LTC donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm
          BTC donation address: 16hJF5mceSojnTD3ZTUDqdRhDyPJzoRakM
          YAC donation address: Y87sptDEcpLkLeAuex6qZioDbvy1qXZEj4
[2015-08-09 16:41:42] NVML GPU monitoring is not available.
[2015-08-09 16:41:42] Starting Stratum on stratum+tcp://ca.simplemulti.com:3433
[2015-08-09 16:41:42] 1 miner threads started, using 'scrypt' algorithm.
[2015-08-09 16:41:42] GPU #0: GeForce GTX 850M with compute capability 5.0
[2015-08-09 16:41:42] GPU #0: interactive: 1, tex-cache: 0 , single-alloc: 0
[2015-08-09 16:41:42] GPU #0: 32 hashes / 4.0 MB per warp.
[2015-08-09 16:41:43] GPU #0: Performing auto-tuning (Patience...)
[2015-08-09 16:41:43] GPU #0: maximum total warps (BxW): 366
[2015-08-09 16:41:59] Stratum detected new block
[2015-08-09 16:42:15] GPU #0: 104654.97 hash/s with configuration T9x20
[2015-08-09 16:42:15] GPU #0: using launch configuration T9x20
[2015-08-09 16:42:15] GPU #0: GeForce GTX 850M, 95.71 khash/s

[sp_]

Try to add -l 5x24


ccminer -a scrypt -l 5x24 ....

in the command line

[pallas]

Here is my contribution to the ccminer-sp effort.

I just posted a pull request:

Code:

Enhanced neoscrypt kernel:

improved hashrate of about 10% (gtx 970)
code cleanup
hardware error printing
other enhancements

Please note that it's not just an hashrate increase, there is a lot of code enhancement and cleanups under the hood.
Also note that I didn't test it on cards other than 970.

Donation BTC address is in my signature.
(Depending on how my effort is supported, I'll see whether to post my future works or not)

Enjoy! :-)

[facanferff]

Try to add -l 5x24


ccminer -a scrypt -l 5x24 ....

in the command line

It seems it is working now:

[2015-08-09 17:00:33] Starting Stratum on stratum+tcp://ca.simplemulti.com:3434
[2015-08-09 17:00:33] NVAPI GPU monitoring enabled.
[2015-08-09 17:00:33] 1 miner thread started, using 'scrypt' algorithm.
[2015-08-09 17:00:33] Binding thread 0 to cpu 0 (mask 1)
[2015-08-09 17:00:34] Stratum difficulty set to 192
[2015-08-09 17:00:34] GPU #0: GeForce GTX 850M with SM 5.0
[2015-08-09 17:00:34] GPU #0: interactive: 0, tex-cache: 0, single-alloc: 0
[2015-08-09 17:00:34] GPU #0: 32 hashes / 4.0 MB per warp.
[2015-08-09 17:00:34] GPU #0: using launch configuration T5x24
[2015-08-09 17:00:34] scrypt factor set to 9 (1024)
[2015-08-09 17:01:08] ca.simplemulti.com:3434 scrypt block 830752
[2015-08-09 17:03:07] ca.simplemulti.com:3434 scrypt block 830753
[2015-08-09 17:06:54] Stratum difficulty set to 8
[2015-08-09 17:10:05] ca.simplemulti.com:3434 scrypt block 830754
[2015-08-09 17:10:14] ca.simplemulti.com:3434 scrypt block 830755
[2015-08-09 17:11:45] ca.simplemulti.com:3434 scrypt block 830756

Thanks very much .

Unfortunately it doesn't show the speed of processing...

[go6ooo1212]

release 58 gained up to 500KH/s more on the rig of six 750ti in quark...

[Zels]

Rlz 58:
~ +100kh on quark for a 750ti ( 6750kh/s )
~ +15kh on neoscrypt for a 750ti ( 195kh/s )

[rudn]

Hi

it seems that -g 2 Parameter is not working anymore
with release 1.5.58 on GTX 750TI -x11

not necessary anymore?

[bathrobehero]

WINDOWS KILL--

The problem I had when I was attempting to use the YAAMP protocol for auto switching was that an error message from Windows would pop up and interrupt the batch file.  Although I disabled Error Reporting, Windows would still send the alert and interrupt the batch file.  Manual interaction was required, mining stopped.

Currently, I do everything manually.  I left-click the red X at the top right of the window, and close ccminer.  If I use ctrl-c, there is a chance of the degradation problem.         --scryptr

The popup error comes from werfault.exe (windows error reporting) so having an independent bat loop killing it every few seconds solves that issue:

:start
title werfault killer
cls
timeout -t 3
taskkill -t -f /im werfault.exe
goto start

[induktor]

quark is faster on the 750ti as well.  (from release 57 to 58 on One of my test rigs the gain is 70khash per card.)

The current default intensity for the gtx750ti is
-i 5 and for  compute 5.2  -i 22

Please help me find the best intensity settings people

Hi sp_, i am testing your new 1558 version, here are my QUARK results:
rig: AM1M motherboard with APU 5350 (25W TDP) 3x gigabyte GTX 970 G1 stock speeds, 14.04 LTS (flash drive), cuda 6.5.19 drv 346.59

default intensity is best BY FAR!, so you NAILED IT

I tested intensities from 20 to 24, all gives me at least 500Kh less, per card, than default.
at intensity 20 is when weird things starts to happen, power consumption SKYROCKETED!, from 650W to 780W   WTF!
at intensity 20 and 21 memory usage reported by nvidia-smi is about 200Mib which is very odd!, at default intensity (not specifiying it) i get 1433 Mib usage per card, at intensities of 23 and 24 memory usage drops again but to 650Mib per card.
at 23 and 24 speed reduces from 16400 per card to around 14000.
another weird thing is that, if i specify -i 22 i get about 500kh less per card than if i don't specify anything, and the memory usage also drops from 1433 to 650Mib.

also at default intensity i get the usual spiky power consumption behavior always had with the GTX 970 about one tick every 800ms, but at lower intensities the power consumption is more constant. (I have a bar graph (vu meter like) power meter i made to track things like this i can do a video of different intensities and power consumption behavior if you like).

Rig: AM1M motherboard with APU 5350 (25 TDP) 3x Gigabyte 750Ti stock speed 14.04LTS (flash drive), cuda 6.5.19 drv 346.59

I found no difference with previous release 1557 in performance and power usage.
if you need me to test something, let me know, happy to help.

[sp_]

Here is my contribution to the ccminer-sp effort.
I just posted a pull request:

Code:

Enhanced neoscrypt kernel:
improved hashrate of about 10% (gtx 970)
code cleanup
hardware error printing
other enhancements

Please note that it's not just an hashrate increase, there is a lot of code enhancement and cleanups under the hood.
Also note that I didn't test it on cards other than 970.
Donation BTC address is in my signature.
(Depending on how my effort is supported, I'll see whether to post my future works or not)
Enjoy! :-)

Thank you for your pullrequest. My gigabyte 970 went from 520-540 to 560-570KHASH. The 750ti's seems to be up 5KHASH, but I think will tune the kernals when I get time.

[sp_]

I submitted another 100KHASH in quark.

the 970 is passing 16,6 MHASH now on standard clocks.

[sp_]

I tested intensities from 20 to 24, all gives me at least 500Kh less, per card, than default.
at intensity 20 is when weird things starts to happen, power consumption SKYROCKETED!, from 650W to 780W   WTF!

Thanks for testing, I found a bug in the intensity code and will submitt a fix. Then I will build release 59 with pallas speedups in neoscrypt.

[sp_]

-Merged pallas pullrequest. (neoscrypt 10% faster on compute 5.2)
-Fixed the intensity parameter
-Submitted speedups in quark,x11,x13,x14 and x15
-Added a new parameter called Xintensity.
  -X can be a number between 0-999

f.eks -X 16

1.5.59(sp-MOD) is available here: (10-08-2015)

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

The sourcecode is available here:

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

[induktor]

Hi sp_
i just downloaded the new version, still says 1558, and the weird thing is that it does not log to syslog like all the others, as a result I have no way of updating display and webpage of the miner, how can I do to log to syslog like the other ones?
thanks!

[skunk]

-Merged pallas pullrequest. (neoscrypt 10% faster on compute 5.2)
-Fixed the intensity parameter
-Submitted speedups in quark,x11,x13,x14 and x15
-Added a new parameter called Xintensity.
  -X can be a number between 0-999

f.eks -X 16

1.5.59(sp-MOD) is available here: (10-08-2015)

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

The sourcecode is available here:

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

what's the difference between -i and -X?
did you notice on api's "threads" command the intensity (I) value always stays at 0.0?

[sp_]

Submitted a new fix for the api and stats:

-The API will only report stats if enabled with the -b switch

Updated the exe file.

[sp_]

-Merged pallas pullrequest. (neoscrypt 10% faster on compute 5.2)
-Fixed the intensity parameter
-Submitted speedups in quark,x11,x13,x14 and x15
-Added a new parameter called Xintensity.
  -X can be a number between 0-999
f.eks -X 16
1.5.59(sp-MOD) is available here: (10-08-2015)
https://github.com/sp-hash/ccminer/releases/
The sourcecode is available here:
https://github.com/sp-hash/ccminer

what's the difference between -i and -X?
did you notice on api's "threads" command the intensity (I) value always stays at 0.0?

the -X switch is just threadcount of X * 256*256. Linary threadcount and not exponential.

-X 1  is the same as -i 16

[scryptr]

RELEASE dot 59 NEOSCRYPT--

Per card hash rates increased measurably:

     2GB GTX 750ti FTW:     +22-25 kh/s per card  (~190kh/s)

     2GB GTX 960 SSC:       +30 kh/s per card  (~345kh/s)

     4GB GTX 970 FTW+      +50-55 kh/s per card  (~550kh/s)

Quark values look slightly better; I haven't mined other algos.       --scryptr