CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 982.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Submitted another small but visible optimalization in quark and x11.

Precalculated 0.75 rounds of blake-512

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: sp_ on August 24, 2015, 01:17:45 PM

Quote from: scryptr on August 24, 2015, 10:53:19 AM

EDIT: I experimented with "-g 2" mining lyra2v2. I had to back the intensity down to "-i 15" from "-i 19.5" in order to launch. I used my GTX 750ti FTW rig, and got matching TIDs on all 12 gpu threads. That is, gpu0 had a TID that always matched gpu0, and so on for gpu9, etc. I received a high number of "Does not validate on CPU!" errors, and went back to mining without it. I did get consistent reports for all 12 virtual gpus, even though only 6 physical gpus are in the rig.
The threads matched, all "gpus" had a report string. Total hash was less, errors were high. --scryptr

The g parameter needs a different kernal. We fixed it for x11,x13,x14,x15,nist5, quark. (blake 80) but the other kernals need more work.

it would probably be easier to use cudastream

Grout

full member

Activity: 139

Merit: 100

Looks like I missed the r61 announcement. Thanks SP! Beer coming your way:
2fe1634ed75e08b0df1deb85bbe98cf8443d4888091fe62086bc560a53c6f690

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: scryptr on August 24, 2015, 10:53:19 AM

EDIT: I experimented with "-g 2" mining lyra2v2. I had to back the intensity down to "-i 15" from "-i 19.5" in order to launch. I used my GTX 750ti FTW rig, and got matching TIDs on all 12 gpu threads. That is, gpu0 had a TID that always matched gpu0, and so on for gpu9, etc. I received a high number of "Does not validate on CPU!" errors, and went back to mining without it. I did get consistent reports for all 12 virtual gpus, even though only 6 physical gpus are in the rig.
The threads matched, all "gpus" had a report string. Total hash was less, errors were high. --scryptr

The g parameter needs a different kernal. We fixed it for x11,x13,x14,x15,nist5, quark. (blake 80) but the other kernals need more work.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

I have reverted most of nelsons changes. Lyrav2 and lyra is faster because i submitted optimalizations yesterday. Mostly on compute 5.2 devices

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: sp_ on August 24, 2015, 10:21:12 AM

Look's like the colors are gone after the pull request from mr Nelson on windows. Might do a rollback

THREAD ID (TID)---

Maybe the TID stats will give you a grip on the gpu-threads switch? My Linux build of commit 958 looks nice, I like the stats, and it produces more hash than release dot 61 (fixed) when mining Lyra2v2. --scryptr

EDIT: I experimented with "-g 2" mining lyra2v2. I had to back the intensity down to "-i 15" from "-i 19.5" in order to launch. I used my GTX 750ti FTW rig, and got matching TIDs on all 12 gpu threads. That is, gpu0 had a TID that always matched gpu0, and so on for gpu9, etc. I received a high number of "Does not validate on CPU!" errors, and went back to mining without it. I did get consistent reports for all 12 virtual gpus, even though only 6 physical gpus are in the rig.

The threads matched, all "gpus" had a report string. Total hash was less, errors were high. --scryptr

zTheWolfz

full member

Activity: 231

Merit: 150

Quote from: djm34 on August 24, 2015, 08:05:32 AM

Quote from: sp_ on August 24, 2015, 08:02:16 AM

Quote from: lawrencelyl on August 24, 2015, 07:45:37 AM

Quote from: djm34 on August 24, 2015, 07:43:20 AM

Quote from: lawrencelyl on August 24, 2015, 07:42:37 AM

Quote from: djm34 on August 24, 2015, 07:39:19 AM

+150kh/s on 750ti
+3MH/s on 980 Grin

Which algo? On latest sp_'s github? Shocked

lyra2rev2, not released yet...

Eagerly waiting Grin

I have only added 500KHASH on the 980 (sp-mod 61-git) Which algos did you rewrite? And will you make it public?

rewrote lyra2 code:

Very nice!
I could test for you on GTX960? ;-)

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Look's like the colors are gone after the pull request from mr Nelson on windows. Might do a rollback

fenomenhaa

sr. member

Activity: 248

Merit: 250

Quote from: sp_ on August 24, 2015, 08:39:22 AM

But I am more interested in your optimized quark kernal, do you mind to share it with us?

Agree that idea. As i said many many times,i'm always ready for donations for improved kernels.Is there anybody have an experience with new maxwell cards(950) ? )))

scryptr

legendary

Activity: 1797

Merit: 1028

CCMINER CONTRIBUTIONS--

I have been donating whenever a significant improvement in CCminer has been made. The last few weeks have been really good, improvement wise. Any contribution by the talented developers and coders should be matched by donations from the miners. --scryptr

lawrencelyl

member

Activity: 94

Merit: 10

Quote from: djm34 on August 24, 2015, 08:30:36 AM

Quote from: sp_ on August 24, 2015, 08:14:46 AM

Quote from: djm34 on August 24, 2015, 08:05:32 AM

Quote from: sp_ on August 24, 2015, 08:02:16 AM

I have only added 500KHASH on the 980 (sp-mod 61-git) Which algos did you rewrite? And will you make it public?

rewrote lyra2 code:

Good job.
I guess the new kernal is using more power?

yes, the 980 runs at 110% tpd and the 750ti close to 95% (on my gainward stock).
Honestly I will try to backport (if possible... not entirely obvious though, the new kernel relies on warp shuffle technique) some of the change to amd before releasing it, I don't want to create more imbalance...

Will it be available for public?
Donation will follow Wink

bensam1231

legendary

Activity: 1764

Merit: 1024

Quote from: bathrobehero on August 24, 2015, 05:30:48 AM

Quote from: sp_ on August 24, 2015, 02:41:33 AM

Quote from: chrysophylax on August 24, 2015, 02:31:32 AM

hi sp ...

initial results of fresh compile and mining at givemecoins using only -X 15 and --diff .5 with no oc or firmware tweaking ... just stock ...

*** ccminer 1.5.61-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
   Built with the 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.

[2015-08-24 16:56:18] tid(0x0000047e) XIntensity set to 15, 983040 cuda threads
[2015-08-24 16:56:23] tid(0x00000515) Starting Stratum on stratum+tcp://mining-ace.granitecoin.com:6001/

[2015-08-24 16:57:55] tid(0x00000515) accepted: 358/360 (99.44%), 24145 kH/s yes!
[2015-08-24 16:57:55] tid(0x00000515) accepted: 359/361 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:55] tid(0x00000515) accepted: 360/362 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:55] tid(0x00000515) Stratum difficulty set to 4
[2015-08-24 16:57:56] tid(0x00000519) GPU #2: GeForce GTX 750 Ti, 4068 Temp= 59C Fan= 47%
[2015-08-24 16:57:56] tid(0x00000515) accepted: 361/363 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:56] tid(0x0000051a) GPU #3: GeForce GTX 750 Ti, 4001 Temp= 59C Fan= 46%
[2015-08-24 16:57:56] tid(0x0000051b) GPU #4: GeForce GTX 750 Ti, 4080 Temp= 62C Fan= 49%
[2015-08-24 16:57:56] tid(0x00000518) GPU #1: GeForce GTX 750 Ti, 3989 Temp= 57C Fan= 45%
[2015-08-24 16:57:56] tid(0x00000515) accepted: 362/364 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:56] tid(0x00000515) accepted: 363/365 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:56] tid(0x00000515) accepted: 364/366 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:56] tid(0x00000515) accepted: 365/367 (99.46%), 24145 kH/s yes!
[2015-08-24 16:57:57] tid(0x00000517) GPU #0: GeForce GTX 750 Ti, 4010 Temp= 61C Fan= 48%
[2015-08-24 16:57:57] tid(0x00000515) accepted: 366/368 (99.46%), 24145 kH/s yes!
[2015-08-24 16:57:58] tid(0x00000519) GPU #2: GeForce GTX 750 Ti, 4092 Temp= 60C Fan= 47%
[2015-08-24 16:57:58] tid(0x00000515) accepted: 367/369 (99.46%), 24146 kH/s yes!
[2015-08-24 16:58:01] tid(0x0000051c) GPU #5: GeForce GTX 750 Ti, 4012 Temp= 61C Fan= 48%
[2015-08-24 16:58:01] tid(0x00000515) accepted: 368/370 (99.46%), 24146 kH/s yes!

when the mining stabalizes in the next 15-30mins - that will show a more accurate figure ...
i like the temp and fan displays too - nice

... why not shorten T=xx F=xx ? ...
what is this new part also - tid? ...
#crysx

Thread id. I think I should remove it. The temps and fanspeed is only working in the linux version. The changes in the output is written by t-nelson, and I have merged his changes into my fork.

I know it's a dead horse by now, but have you tried having the first few lines of ccminer fixed instead of scrolling and put some info like temp and fan speed there, like sg/cgminer? The issue with the first approach was that linux wouldn't support it but surely there's another way.

OOOOH SGMINER STATS! *starts fapping vigorously*

Everyone is overly concerned about Nix, but pretty sure most people don't mine with it. Maybe ridiculously huge farms use it (I know a few that don't), but I doubt that's the main 'clientele' not that there is really any clientele when you're making something for free though.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

But I am more interested in your optimized quark kernal, do you mind to share it with us?

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: sp_ on August 24, 2015, 08:14:46 AM

Quote from: djm34 on August 24, 2015, 08:05:32 AM

Quote from: sp_ on August 24, 2015, 08:02:16 AM

I have only added 500KHASH on the 980 (sp-mod 61-git) Which algos did you rewrite? And will you make it public?

rewrote lyra2 code:

Good job.
I guess the new kernal is using more power?

yes, the 980 runs at 110% tpd and the 750ti close to 95% (on my gainward stock).
Honestly I will try to backport (if possible... not entirely obvious though, the new kernel relies on warp shuffle technique) some of the change to amd before releasing it, I don't want to create more imbalance...

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: djm34 on August 24, 2015, 08:05:32 AM

Quote from: sp_ on August 24, 2015, 08:02:16 AM

I have only added 500KHASH on the 980 (sp-mod 61-git) Which algos did you rewrite? And will you make it public?

rewrote lyra2 code:

Good job.
I guess the new kernal is using more power?

chrysophylax

legendary

Activity: 2940

Merit: 1091

--- ChainWorks Industries ---

Quote from: sp_ on August 24, 2015, 04:28:43 AM

Quote from: chrysophylax on August 24, 2015, 03:05:37 AM

and temp / fan in linux only? ... that would be a first for linux to have wouldnt it? ... nice merge in my opinion ... Wink

...
#crysx

It's in the windows version as well if you build 64bit and not x86.
The reason is that NVIDIA has removed the support for the API calls in the 32bit application.
On windows the 64bit builds are hashing slower than the 32 bit builds.

i was playing with a little humour there sp ...

usually its linux thats well behind windows based advances that nvidia bring out ...

but i understand why now ...

tanx ...

#crysx

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: sp_ on August 24, 2015, 08:02:16 AM

Quote from: lawrencelyl on August 24, 2015, 07:45:37 AM

Quote from: djm34 on August 24, 2015, 07:43:20 AM

Quote from: lawrencelyl on August 24, 2015, 07:42:37 AM

Quote from: djm34 on August 24, 2015, 07:39:19 AM

+150kh/s on 750ti
+3MH/s on 980 Grin

Which algo? On latest sp_'s github? Shocked

lyra2rev2, not released yet...

Eagerly waiting Grin

I have only added 500KHASH on the 980 (sp-mod 61-git) Which algos did you rewrite? And will you make it public?

rewrote lyra2 code:

chrysophylax

legendary

Activity: 2940

Merit: 1091

--- ChainWorks Industries ---

Quote from: djm34 on August 24, 2015, 07:43:20 AM

Quote from: lawrencelyl on August 24, 2015, 07:42:37 AM

Quote from: djm34 on August 24, 2015, 07:39:19 AM

+150kh/s on 750ti
+3MH/s on 980 Grin

Which algo? On latest sp_'s github? Shocked

lyra2rev2, not released yet...

id pay for that - if its private djm34? ...

150KH increase is nice on a 750ti ...

#crysx

chrysophylax

legendary

Activity: 2940

Merit: 1091

--- ChainWorks Industries ---

Quote from: bathrobehero on August 24, 2015, 05:30:48 AM

Quote from: sp_ on August 24, 2015, 02:41:33 AM

Quote from: chrysophylax on August 24, 2015, 02:31:32 AM

hi sp ...

initial results of fresh compile and mining at givemecoins using only -X 15 and --diff .5 with no oc or firmware tweaking ... just stock ...

*** ccminer 1.5.61-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
   Built with the 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.

[2015-08-24 16:56:18] tid(0x0000047e) XIntensity set to 15, 983040 cuda threads
[2015-08-24 16:56:23] tid(0x00000515) Starting Stratum on stratum+tcp://mining-ace.granitecoin.com:6001/

[2015-08-24 16:57:55] tid(0x00000515) accepted: 358/360 (99.44%), 24145 kH/s yes!
[2015-08-24 16:57:55] tid(0x00000515) accepted: 359/361 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:55] tid(0x00000515) accepted: 360/362 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:55] tid(0x00000515) Stratum difficulty set to 4
[2015-08-24 16:57:56] tid(0x00000519) GPU #2: GeForce GTX 750 Ti, 4068 Temp= 59C Fan= 47%
[2015-08-24 16:57:56] tid(0x00000515) accepted: 361/363 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:56] tid(0x0000051a) GPU #3: GeForce GTX 750 Ti, 4001 Temp= 59C Fan= 46%
[2015-08-24 16:57:56] tid(0x0000051b) GPU #4: GeForce GTX 750 Ti, 4080 Temp= 62C Fan= 49%
[2015-08-24 16:57:56] tid(0x00000518) GPU #1: GeForce GTX 750 Ti, 3989 Temp= 57C Fan= 45%
[2015-08-24 16:57:56] tid(0x00000515) accepted: 362/364 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:56] tid(0x00000515) accepted: 363/365 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:56] tid(0x00000515) accepted: 364/366 (99.45%), 24145 kH/s yes!
[2015-08-24 16:57:56] tid(0x00000515) accepted: 365/367 (99.46%), 24145 kH/s yes!
[2015-08-24 16:57:57] tid(0x00000517) GPU #0: GeForce GTX 750 Ti, 4010 Temp= 61C Fan= 48%
[2015-08-24 16:57:57] tid(0x00000515) accepted: 366/368 (99.46%), 24145 kH/s yes!
[2015-08-24 16:57:58] tid(0x00000519) GPU #2: GeForce GTX 750 Ti, 4092 Temp= 60C Fan= 47%
[2015-08-24 16:57:58] tid(0x00000515) accepted: 367/369 (99.46%), 24146 kH/s yes!
[2015-08-24 16:58:01] tid(0x0000051c) GPU #5: GeForce GTX 750 Ti, 4012 Temp= 61C Fan= 48%
[2015-08-24 16:58:01] tid(0x00000515) accepted: 368/370 (99.46%), 24146 kH/s yes!

when the mining stabalizes in the next 15-30mins - that will show a more accurate figure ...
i like the temp and fan displays too - nice

... why not shorten T=xx F=xx ? ...
what is this new part also - tid? ...
#crysx

Thread id. I think I should remove it. The temps and fanspeed is only working in the linux version. The changes in the output is written by t-nelson, and I have merged his changes into my fork.

I know it's a dead horse by now, but have you tried having the first few lines of ccminer fixed instead of scrolling and put some info like temp and fan speed there, like sg/cgminer? The issue with the first approach was that linux wouldn't support it but surely there's another way.

i have suggested these requests a very long time ago ... as in the 'stats candy' of sgminer but the performance of ccminer in one package ...

i think it will require a full rebuild of ccminer if it comes to that - maybe? ...

i like the way sgminer prints stats and scrolls the active stats below the cumulative / current stats ...

no one has come to the party with this - and i think it is due to a full rewrite of ccminer ...

o well - we can hope

...

#crysx

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer