CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1045.

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: scryptr on July 07, 2015, 11:14:55 PM

DJM34, SP_ --

I flipped you each a nickle. Thank you for your hard work! I hope my 960's will be able to mine Lyra2 on Windows at about 1250kh/s soon, maybe more!

Thanks! --scryptr

P.S. I was able to get DJM34's Windows binary to run on my Win 7 x64 system with a 2GB GTX960 SSC with ONLY the performance setting of "-i 16.3". No other performance settings were used. Algo, username, password were as standard.

Result: 1175kh/s mining Lyra2 --scryptr

thanks,
don't forget to run at p0 state using nvidia-smi , that gives the possibility to oc the memclock (it will run also at a somewhat)

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: sp_ on July 08, 2015, 01:43:44 AM

Quote from: scryptr on July 07, 2015, 09:11:42 PM

On my Linux boxes, SP_'s build 843 compiled and mines Lyra2 at 1850kh/s on my 970 FTW+ cards, and at 1050kh/s on my 750ti FTW cards. The 750ti FTW cards were running at 825kh/s on the SP_'s release dot 50.

Did you try after my latest commit? I get 40khash + on my gigabyte windforce cards with a 6pins connector.(750ti) The compute 5.2 cards are unchanged as they use another kernal.

Here is the commit:

https://github.com/sp-hash/ccminer/commit/384d4cc461d38fdfb2243cb806806cdccad98074

The commit is not big but it reduces the register usage from 185 to 113. and reduces the codesize wich gives less pressure on the instructioncache.
(less memory usage)

the pragma unroll were chosen with care and they enhance the hashrate by about that same amount on one of my card (most likely the 980, that might decrease the hashrate on the 900 serie...)

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: CapnBDL on July 07, 2015, 11:30:30 PM

Not lookin' to start a war. Made those comments so that maybe noobs wouldn't make mistakes. On a development thread like this...sooo many will be confused. That's all. Comments need to be FULLY explained (ie. if you are part of the CUDA team, let us know, etc), otherwise noobs will think they may be missing an upgrade (& it's not even released yet. Heck, 7 is questionable/buggy.).

And a 'test pool'...really, and you don't mention that to start with? I hope you own lots of DASH, cuz I want my cut!! Kiddin', but could have ended badly.

You are the one sounding a bit like a noob actually... (no one will do his presentation every day, it is up to you to know who is who Grin

)

regarding cuda 7.5, there are a bit of good and a bit of bad...
on lyra2re, there is a clear +100kh/s on the 980 and +60 on the 750ti (from 1140 to 1200kh/s on my card oc at +150/+150) and also a clear -700kh/s on the 780ti and it is worst on the new neoscrypt code (unpublished) even though some aspect are better...

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: scryptr on July 07, 2015, 09:11:42 PM

On my Linux boxes, SP_'s build 843 compiled and mines Lyra2 at 1850kh/s on my 970 FTW+ cards, and at 1050kh/s on my 750ti FTW cards. The 750ti FTW cards were running at 825kh/s on the SP_'s release dot 50.

Did you try after my latest commit? I get 40khash + on my gigabyte windforce cards with a 6pins connector.(750ti) The compute 5.2 cards are unchanged as they use another kernal.

Here is the commit:

https://github.com/sp-hash/ccminer/commit/384d4cc461d38fdfb2243cb806806cdccad98074

The commit is not big but it reduces the register usage from 185 to 113. and reduces the codesize wich gives less pressure on the instructioncache.
(less memory usage)

CapnBDL

hero member

Activity: 1064

Merit: 500

MOBU

Not lookin' to start a war. Made those comments so that maybe noobs wouldn't make mistakes. On a development thread like this...sooo many will be confused. That's all. Comments need to be FULLY explained (ie. if you are part of the CUDA team, let us know, etc), otherwise noobs will think they may be missing an upgrade (& it's not even released yet. Heck, 7 is questionable/buggy.).

And a 'test pool'...really, and you don't mention that to start with? I hope you own lots of DASH, cuz I want my cut!! Kiddin', but could have ended badly.

scryptr

legendary

Activity: 1797

Merit: 1028

DJM34, SP_ --

I flipped you each a nickle. Thank you for your hard work! I hope my 960's will be able to mine Lyra2 on Windows at about 1250kh/s soon, maybe more!

Thanks! --scryptr

P.S. I was able to get DJM34's Windows binary to run on my Win 7 x64 system with a 2GB GTX960 SSC with ONLY the performance setting of "-i 16.3". No other performance settings were used. Algo, username, password were as standard.

Result: 1175kh/s mining Lyra2 --scryptr

antonio8

legendary

Activity: 1400

Merit: 1000

Quote from: CapnBDL on July 07, 2015, 10:45:28 PM

Quote from: Epsylon3 on July 07, 2015, 08:52:17 PM

yiimp is a "test pool" i try to set up... without auto exchange, i will update the main page to explain better soon...

will not be like the yaamp multipool system which require a lot of attention about trades

else... CUDA 7.5 really improve ccminer, on almost all algos :p

I don't know how you can say CUDA7.5 does so much better. It was just put out to developers, hence the 'RC' designation. Stands for 'Release Candidate', meaning it's in the early stages.

Concerning your 'test pool'; I wouldn't broadcast you are trying this until you are ready to pay-up! I almost started mining there thinking I would be paid for the work I was doing. Just sayin'!

CapnBDL

hero member

Activity: 1064

Merit: 500

MOBU

Quote from: Epsylon3 on July 07, 2015, 08:52:17 PM

yiimp is a "test pool" i try to set up... without auto exchange, i will update the main page to explain better soon...

will not be like the yaamp multipool system which require a lot of attention about trades

else... CUDA 7.5 really improve ccminer, on almost all algos :p

I don't know how you can say CUDA7.5 does so much better. It was just put out to developers, hence the 'RC' designation. Stands for 'Release Candidate', meaning it's in the early stages.

Concerning your 'test pool'; I wouldn't broadcast you are trying this until you are ready to pay-up! I almost started mining there thinking I would be paid for the work I was doing. Just sayin'!

scryptr

legendary

Activity: 1797

Merit: 1028

DJM34 and LYRA2 --

I was able to get DJM34's Windows binary to run on my Win 8 x64 rig. This rig has 5x750ti SSC and 1x960 4GB FTW.

Overall, the rig is faster by almost 2Mhash. My 750ti cards run at 1050kh/s each, and the 4GB 960 FTW gets 1175kh/s while mining Lyra2. This is the first time the 960 has run faster than the 750ti cards. Formerly, it ran at 500kh/s (+/- 75kh/s), compared to 725kh/s for the 750ti cards.

On my Win 7 x64 box, where my 2gb 960 SSC is the only graphics card, the DJM34 binary won't launch, as I stated on the previous page.

On my Linux boxes, SP_'s build 843 compiled and mines Lyra2 at 1850kh/s on my 970 FTW+ cards, and at 1050kh/s on my 750ti FTW cards. The 750ti FTW cards were running at 825kh/s on the SP_'s release dot 50.

--scryptr

Epsylon3

legendary

Activity: 1484

Merit: 1082

ccminer/cpuminer developer

yiimp is a "test pool" i try to set up... without auto exchange, i will update the main page to explain better soon...

will not be like the yaamp multipool system which require a lot of attention about trades

else... CUDA 7.5 really improve ccminer, on almost all algos :p

flipclip

member

Activity: 111

Merit: 10

So now there is yiimp?

http://yiimp.ccminer.org/

CapnBDL

hero member

Activity: 1064

Merit: 500

MOBU

FYI: CUDA 7.5RC just out.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

submitted a 40khash increase in lyra2 on the 750ti only. (3.5%)

(reduced the register usage from 185 to 113.)

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: djm34 on July 07, 2015, 01:57:41 PM

Quote from: sp_ on July 07, 2015, 12:58:45 PM

My gtx 970 is up 60% in lyra nice.
did you change some other kernals as well? I just merged the lyra kernal and kept the others..

cuda_vector.h might have changed as well
btw you also need to change opt_scantime in miner_thread and put it to 60 (or something like that) gives more stable mining in many algo

@SP_, for Lyra2 Merge--

Please get it to a Windows release. I think it is time for a new round of donations!

@DJM34-- I keep getting "Cuda error in func 'scanhash_Lyra2' at line 93 : out of memory.", on launch. This is on NiceHash with "--diff 1", using a 2GB GTX 960 SSC on Win 7 x64, and your latest binary. The Win 7 system has 8GB of memory.

Thanks! --scryptr

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: sp_ on July 07, 2015, 12:58:45 PM

My gtx 970 is up 60% in lyra nice.
did you change some other kernals as well? I just merged the lyra kernal and kept the others..

cuda_vector.h might have changed as well
btw you also need to change opt_scantime in miner_thread and put it to 60 (or something like that) gives more stable mining in many algo

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

My gtx 970 is up 60% in lyra nice.
did you change some other kernals as well? I just merged the lyra kernal and kept the others..

bensam1231

legendary

Activity: 1764

Merit: 1024

Quote from: djm34 on July 07, 2015, 12:32:44 PM

Quote from: sp_ on July 07, 2015, 11:47:25 AM

Quote from: djm34 on July 07, 2015, 07:57:58 AM

fast(er) lyra algo has been released:

windows binaries: https://mega.co.nz/#!5ZkVmLZb!qgTTz4OJFsAtJNLqwYqf6mGvaSWwHiVMBoDUlOk1sDc
source code: https://github.com/djm34/ccminer-lyra

1140kh/s on 750ti
2500kh/s on 980
2900kh/s on 780ti (and 3.3MH/s if you are lucky enough to screw your compilation... got that once, mostly random Grin

)

Just added some of the modules to my fork. Lyra is more than 40% faster.

Nicely done by the Cuda expert djm34

thanks, it goes up to 90% faster for the big guns

Appreciate it... Hopefully Lyra will have a upturn again.

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: sp_ on July 07, 2015, 11:47:25 AM

Quote from: djm34 on July 07, 2015, 07:57:58 AM

fast(er) lyra algo has been released:

windows binaries: https://mega.co.nz/#!5ZkVmLZb!qgTTz4OJFsAtJNLqwYqf6mGvaSWwHiVMBoDUlOk1sDc
source code: https://github.com/djm34/ccminer-lyra

1140kh/s on 750ti
2500kh/s on 980
2900kh/s on 780ti (and 3.3MH/s if you are lucky enough to screw your compilation... got that once, mostly random Grin

)

Just added some of the modules to my fork. Lyra is more than 40% faster.

Nicely done by the Cuda expert djm34

thanks, it goes up to 90% faster for the big guns

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: djm34 on July 07, 2015, 07:57:58 AM

fast(er) lyra algo has been released:

windows binaries: https://mega.co.nz/#!5ZkVmLZb!qgTTz4OJFsAtJNLqwYqf6mGvaSWwHiVMBoDUlOk1sDc
source code: https://github.com/djm34/ccminer-lyra

1140kh/s on 750ti
2500kh/s on 980
2900kh/s on 780ti (and 3.3MH/s if you are lucky enough to screw your compilation... got that once, mostly random Grin

)

Just added some of the modules to my fork. Lyra is more than 40% faster.

Nicely done by the Cuda expert djm34

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Use ccminer sp-mod release 53. Run with -g 2. should be closer to 3,2MHASH. But it uses 2x the memory and a little more power around 45-50watt.

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1045. (Read 2347601 times)