[ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX]

[tc61]

Attached is nvMiner or otherwise known as ccminer 1.2U-a (U=unified)

I renamed it from ccminer to nvminer to make it clearer if it's a Christian release or from 3rd party.

This includes the following algos (unified from multiple devs):
heavy
mjollnir
fugue256
groestl
myr-gr
dmd-gr
jackpot
quark
anime
qubit
nist5
cryptonight
x11
x13
x14
x15

Also includes a few subtle changes and a couple of speedups to some routines.

https://ayarscloud.tonidoid.com/urljfz5fy

Have fun,
Carlo

Carlo, thank you very much!
Posted from Bitcointa.lk - #3zxWN4eLUtjDc8aZ

[Amph]

Attached is nvMiner or otherwise known as ccminer 1.2U-a (U=unified)

I renamed it from ccminer to nvminer to make it clearer if it's a Christian release or from 3rd party.

This includes the following algos (unified from multiple devs)...................
Also includes a few subtle changes and a couple of speedups to some routines.

https://ayarscloud.tonidoid.com/urljfz5fy

Have fun,
Carlo

Any chance we can take a look at the code for those speedups? Maybe easier to just setup a github

was compiled from this https://github.com/djm34/ccminer, but there will be another release with fresh-algo

compiling now require only 10min, at least for me, really fast

[go6ooo1212]

This version brings me above 1600 MH/s on X15 with my PALIT 750TI OC, nice ))

[Amph]

This version brings me above 1600 MH/s on X15 with my PALIT 750TI OC, nice ))

can you tell me your core/mem settings? with 1300 core i'm getting only 1550

[go6ooo1212]

This version brings me above 1600 MH/s on X15 with my PALIT 750TI OC, nice ))

can you tell me your core/mem settings? with 1300 core i'm getting only 1550

Sure no problem , I have a tig of 6 Palit 750TI OC/ 600W PSU / Asrock H81 Pro BTC / Celeron 1820/ 4GB RAM
Nvidia drv 340.43
MSI Afterburner 3.0.1 Core clock +108 / Mem +108

[Amph]

This version brings me above 1600 MH/s on X15 with my PALIT 750TI OC, nice ))

can you tell me your core/mem settings? with 1300 core i'm getting only 1550

Sure no problem , I have a tig of 6 Palit 750TI OC/ 600W PSU / Asrock H81 Pro BTC / Celeron 1820/ 4GB RAM
Nvidia drv 340.43
MSI Afterburner 3.0.1 Core clock +108 / Mem +108

i think the drivers are making a difference here(still using the whql), because i have core 1340(default) and mem +200

[cayars]

Attached is nvMiner or otherwise known as ccminer 1.2U-a (U=unified)

I renamed it from ccminer to nvminer to make it clearer if it's a Christian release or from 3rd party.

This includes the following algos (unified from multiple devs)...................
Also includes a few subtle changes and a couple of speedups to some routines.

https://ayarscloud.tonidoid.com/urljfz5fy

Have fun,
Carlo

Any chance we can take a look at the code for those speedups? Maybe easier to just setup a github

Sure, but then I'd have to kill ya.

I will post the sources after version B is released which will have some additional functionality. I also want to clean up 3 or 4 more functions first.
Maybe also add overclocking of core and memory as command line arguments too (similar to SGMiner).

BTW, I forgot to mention this does have the new code djm34 just released for the whirlpool routines.

Also  you will notice the output of shares is somewhat different:
[2014-07-04 16:25:53] GPU #1: GeForce GTX 660, 2920 khash/s
[2014-07-04 16:25:53] accepted: 46/47 (97.87%), 6425 kh/s (5960-47) - Yay
[2014-07-04 16:25:56] GPU #1: GeForce GTX 660, 3548 khash/s
[2014-07-04 16:25:56] accepted: 47/48 (97.92%), 7052 kh/s (5982-48) - Yay
[2014-07-04 16:25:57] GPU #1: GeForce GTX 660, 3527 khash/s
[2014-07-04 16:25:57] accepted: 48/49 (97.96%), 7031 kh/s (6004-49) - Yay
removed a few lines
[2014-07-04 16:26:03] accepted: 50/51 (98.04%), 6697 kh/s (6055) - Yay
[2014-07-04 16:26:03] GPU #0: GeForce GTX 660, 3364 khash/s
[2014-07-04 16:26:03] accepted: 51/52 (98.08%), 6389 kh/s (6095) - Yay
[2014-07-04 16:26:04] GPU #0: GeForce GTX 660, 455.99 khash/s

You'll notice right before the YAY there is a number in parenthesis.  Possible a number dash number.
This is the average of the last 50 hashes submitted.  If there is a dash and another number (1 to 50) then the program was just started.  The number after the dash is the number of shares submitted.  So if it's 10 then the "average" is based on 10 shares.  If it's dash 30 then the average is for 30 shares, etc.  Once there is no dash number then you know you have over 50 shares submitted and the average is just based on the last 50 shares submitted.  

You can see the difference in the two bolded ones above.

It's a bit more accurate indication of what to expect on a pool.  For example if you submit a rejected share the hash won't count but it will lower your overall average hash rate.  I "manipulated" the running program on purpose at the start to get some rejected shares to help show this.  So you can see from the results above the "average" is lower than what is traditionally displayed due to the rejects.

The last 50 is typically good enough but an algo like jha (jackpot) is very fast and you will still see the average jump around a bit.

Carlo, thank you very much!
Posted from Bitcointa.lk - #3zxWN4eLUtjDc8aZ

Welcome

Is there a Linux version?

Not at present but after I publish the revised source code anyone could compile it on Linux.

Carlo

[go6ooo1212]

This version brings me above 1600 MH/s on X15 with my PALIT 750TI OC, nice ))

can you tell me your core/mem settings? with 1300 core i'm getting only 1550

Sure no problem , I have a tig of 6 Palit 750TI OC/ 600W PSU / Asrock H81 Pro BTC / Celeron 1820/ 4GB RAM
Nvidia drv 340.43
MSI Afterburner 3.0.1 Core clock +108 / Mem +108

i think the drivers are making a difference here(still using the whql), because i have core 1340(default) and mem +200

half of mine are working at 1320 , other half at 1340 ,but that difference is from extenders I suppose

[StuffOfInterest]

BTW, I forgot to mention this does have the new code djm34 just released for the whirlpool routines.

Also  you will notice the output of shares is somewhat different:
[2014-07-04 16:25:53] GPU #1: GeForce GTX 660, 2920 khash/s
[2014-07-04 16:25:53] accepted: 46/47 (97.87%), 6425 kh/s (5960-47) - Yay
[2014-07-04 16:25:56] GPU #1: GeForce GTX 660, 3548 khash/s
[2014-07-04 16:25:56] accepted: 47/48 (97.92%), 7052 kh/s (5982-48) - Yay
[2014-07-04 16:25:57] GPU #1: GeForce GTX 660, 3527 khash/s
[2014-07-04 16:25:57] accepted: 48/49 (97.96%), 7031 kh/s (6004-49) - Yay
removed a few lines
[2014-07-04 16:26:03] accepted: 50/51 (98.04%), 6697 kh/s (6055) - Yay
[2014-07-04 16:26:03] GPU #0: GeForce GTX 660, 3364 khash/s
[2014-07-04 16:26:03] accepted: 51/52 (98.08%), 6389 kh/s (6095) - Yay
[2014-07-04 16:26:04] GPU #0: GeForce GTX 660, 455.99 khash/s

You'll notice right before the YAY there is a number in parenthesis.  Possible a number dash number.
This is the average of the last 50 hashes submitted.  If there is a dash and another number (1 to 50) then the program was just started.  The number after the dash is the number of shares submitted.  So if it's 10 then the "average" is based on 10 shares.  If it's dash 30 then the average is for 30 shares, etc.  Once there is no dash number then you know you have over 50 shares submitted and the average is just based on the last 50 shares submitted.  

You can see the difference in the two bolded ones above.

It's a bit more accurate indication of what to expect on a pool.  For example if you submit a rejected share the hash won't count but it will lower your overall average hash rate.  I "manipulated" the running program on purpose at the start to get some rejected shares to help show this.  So you can see from the results above the "average" is lower than what is traditionally displayed due to the rejects.

The last 50 is typically good enough but an algo like jha (jackpot) is very fast and you will still see the average jump around a bit.

That's really useful for setting up my auto-switcher.  It was a pain in the butt before to gather a bunch of the numbers manually and average. With this reporting all I have to do is let the miner run for a few minutes and take the last reported number.  Thanks a lot to everyone who contributed code which went into this release.

I've tested this against x11, x13, x15, and Nist5 on NiceHash and had shared accepted for all four.  Also seem to have a (slight) speed bump for all of them but that could be the result of bad estimating on my part for all but x15 which we know is better now.

Now that there are several people making meaningful contributions, I hope an effort will be made to setup a shared (but moderated) repository to allow code submissions from multiple people and new people.  BTW, I wouldn't be upset to see the splitscreen code make it in too if it can be isolated in a way that won't break Linux builds.

I know there is not a lot of interest in some of these algorithms these days, but it would be nice to bring the algorithms over from cudaMiner into nvminer (or ccminer) so we can have one code base to mine them all.

Good work all!

[tc61]

Has anyone make any improvements to the cryptonight algo? Or does it still slow the whole computer down? I would love to mine this but make the computer unusable. thx
Posted from Bitcointa.lk - #ffEIUBMaa2IUxu5Z

[DemosMirak]

Attached is nvMiner or otherwise known as ccminer 1.2U-a (U=unified)

I renamed it from ccminer to nvminer to make it clearer if it's a Christian release or from 3rd party.

This includes the following algos (unified from multiple devs):
heavy
mjollnir
fugue256
groestl
myr-gr
dmd-gr
jackpot
quark
anime
qubit
nist5
cryptonight
x11
x13
x14
x15

Also includes a few subtle changes and a couple of speedups to some routines.

https://ayarscloud.tonidoid.com/urljfz5fy

Have fun,
Carlo

Nice, starting it up remotely on my rigs at home now.

Edit: Perhaps add donation adresses for Cbuchner, tsiv, djm and the other contributors, along with what they contributed. Also mention pooler (and maybe his donation address), as his miner was the basis for this one, I seem to remember.

[dotvibe]

anyone actually compare CUDA 5.5 to CUDA 6.0 compiles? see if there really is a speed difference?

[cayars]

That's really useful for setting up my auto-switcher.  It was a pain in the butt before to gather a bunch of the numbers manually and average. With this reporting all I have to do is let the miner run for a few minutes and take the last reported number.  Thanks a lot to everyone who contributed code which went into this release.

I've tested this against x11, x13, x15, and Nist5 on NiceHash and had shared accepted for all four.  Also seem to have a (slight) speed bump for all of them but that could be the result of bad estimating on my part for all but x15 which we know is better now.

Now that there are several people making meaningful contributions, I hope an effort will be made to setup a shared (but moderated) repository to allow code submissions from multiple people and new people.  BTW, I wouldn't be upset to see the splitscreen code make it in too if it can be isolated in a way that won't break Linux builds.

I thought about this but didn't for exactly the reason you mentioned.

I know there is not a lot of interest in some of these algorithms these days, but it would be nice to bring the algorithms over from cudaMiner into nvminer (or ccminer) so we can have one code base to mine them all.

Good work all!

I have never looked at the code for CudaMiner.  Can anyone comment how hard this will be to do?  I have a feeling it's not all that easy or Chris would have probably done it by now.

Has anyone make any improvements to the cryptonight algo? Or does it still slow the whole computer down? I would love to mine this but make the computer unusable. thx
Posted from Bitcointa.lk - #ffEIUBMaa2IUxu5Z

I'm working on this right now and will have a "B" version up that may help a bit for CryptoNight (not sure yet). Need to test on Kepler yet as well as more on Maxwell.  I'm new to CUDA and don't really know the architecture yet but from what I do know (unless I'm totally off base) the CryptoNight algo aren't optimized for matching the hardware (threads used, block sizes, etc).  I'm playing.

Nice, starting it up remotely on my rigs at home now.

Edit: Perhaps add donation adresses for Cbuchner, tsiv, djm and the other contributors, along with what they contributed. Also mention pooler (and maybe his donation address), as his miner was the basis for this one, I seem to remember.

I'm going to add a new screen similar to the help screen itself with this info along with what each dev contributed. I'm going to PM each dev for a few different coin donation addresses for this page.  This way it will be clear what coin address belongs to who.  Right now I'm not sure what coin address belongs to each of them.

BTW, if you look at the help screen much of what was on the startup is now there plus a link back to this forum.

[RavenXBR]

I have updated github with a new version of the whirlpool algo (x15)
It compiles much faster and it is slightly faster (got rid of make_ulonglong  )
also no more "don't validate on cpu"


750ti: 1.6MHash/s (before 1.4MHash/s
780ti: 3.6MHash/s (2.9MHash/s from the github release or 3.4MHash/s from an exe I put here)

https://github.com/djm34/ccminer

Sweet!

Compiled version?

As usual, here:
http://cryptomining-blog.com/2991-updated-ccminer-fork-with-x15-algorithm-support-now-available/

From 1.3M to 1.5M

Thx to djm34... 

[djm34]

I have updated github with a new version of the whirlpool algo (x15)
It compiles much faster and it is slightly faster (got rid of make_ulonglong  )
also no more "don't validate on cpu"


750ti: 1.6MHash/s (before 1.4MHash/s
780ti: 3.6MHash/s (2.9MHash/s from the github release or 3.4MHash/s from an exe I put here)

https://github.com/djm34/ccminer

Sweet!

Compiled version?

As usual, here:
http://cryptomining-blog.com/2991-updated-ccminer-fork-with-x15-algorithm-support-now-available/

From 1.3M to 1.5M

Thx to djm34...  

you welcome.
Why would I bother compiling anything any more ?  
(in the article they say around 3MHhash/s for the 780ti, it is rather around 3.5~3.6MHash/s (with potentially 3.7, if I can find a way to keep the gpu usage at 98~99%)

[cayars]

Attached is nvMiner or otherwise known as ccminer 1.2U-B (U=unified)

Same as the last version but with possible improvements to CryptoNight
https://ayarscloud.tonidoid.com/urljfz5fy

If you run CryptoNight can you try this version with and without your usual -l and let me know what hash speeds you get?
Also what cards you have?

Have fun,
Carlo

[Equitum]

Attached is nvMiner or otherwise known as ccminer 1.2U-B (U=unified)

Same as the last version but with possible improvements to CryptoNight
https://ayarscloud.tonidoid.com/urljfz5fy

If you run CryptoNight can you try this version with and without your usual -l and let me know what hash speeds you get?
Also what cards you have?

Have fun,
Carlo

Thanks for putting this all together in one package!

For CryptoNight without the launch config, nvminer gave me 64 blocks of 8 thread (8x64), and gave me about 385 h/s on my 780 Ti Classified. -l 4x120 gives me 475 h/s, for reference.

[tbearhere]

Be patient there will be more opportunities. Right now I cant complain I made ROI with the JPC and MIN. Lets start pilling some BTC for those 880.

how many kh/s in scrypt do you think it will do?

[djm34]

Be patient there will be more opportunities. Right now I cant complain I made ROI with the JPC and MIN. Lets start pilling some BTC for those 880.

how many kh/s in scrypt do you think it will do?

not enough for its power usage compared to asic.
based on the 750ti, it should do around 1.5MHash/s

[tbearhere]

Be patient there will be more opportunities. Right now I cant complain I made ROI with the JPC and MIN. Lets start pilling some BTC for those 880.

how many kh/s in scrypt do you think it will do?

not enough for its power usage compared to asic.
based on the 750ti, it should do around 1.5MHash/s

That would be great on a 6 card rig with X11 ect.