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

[bigjme]

If you had X money to spend on buying GPUs.  What would you buy/purchase or would you wait for the next wave of maxwell?

Maybe I wouldn't actually invest in GPUs, considering the uncertainity about making ROI

Christian

Your telling me! £65 in a month, Only need to earn another 6 times that to break even. If only bitcoins value would go back up!

[S_tring]

Well if it depends on the algo, I guess the new version won't fix it... Now the question is if it worth getting a strong CPU now for my rig...

intel i7 3770k@4500MHz 1 instance of ccminer with x11 use 6,5% of cpu. Using 19,5% now for 660ti + 2x 750ti (660ti take same percents of usage cpu as 750ti)

On an i3 750ti uses 9% of CPU, on celeron 19%.
Maybe Christian can give a better answer what causes the high CPU usage.

the "throughput" value was set pretty low (i.e. the number of threads thrown at the GPUs at once). The current github code has it increased 8-fold.
This should lower CPU use.

Great work!
750ti now uses ~1,5% (previously 6,5%) and got 2150 mhash/s (previously 1700 mhash/s)!!
660ti got 2300 mhash/s (previously 1900 mhash/s)

Outstanding work Christian! Can't wait to try all the fancy new x11 coins with nvidia power! Zelante, I was wondering if you would please share the compiled windows version, it's quite late over here and I don't want to deal with VS tonight. Thanks 

Where is this new version, I only see ver 1.0 on the official release (github)?

[cbuchner1]

Where is this new version, I only see ver 1.0 on the official release (github)?

gitub HEAD revision. No release.

[supercoin434]

If a kind soul coul compile for compute 3.0 and 3.5 ,, yourishiku ....

[cbuchner1]

new timeline for ccMiner v1.0 on X11. Note how the table based Groestl completely dominates the runtime.
SIMD has been cut into four pieces, and shrunk to a runtime shorter than Echo already.

You may want to compare this to the timeline I posted earlier where Groestl was optimized ("Killer" version)
but the SIMD hash was particularly slow and nasty.

[zelante]

new timeline for ccMiner v1.0 on X11. Note how the table based Groestl completely dominates the runtime.
SIMD has been cut into four pieces, and shrunk to a runtime shorter than Echo already.

You may want to compare this to the timeline I posted earlier where Groestl was optimized ("Killer" version)
but the SIMD hash was particularly slow and nasty.

but with "killer groestl" this problem is not problem? 

[gpuminer]

Well if it depends on the algo, I guess the new version won't fix it... Now the question is if it worth getting a strong CPU now for my rig...

intel i7 3770k@4500MHz 1 instance of ccminer with x11 use 6,5% of cpu. Using 19,5% now for 660ti + 2x 750ti (660ti take same percents of usage cpu as 750ti)

On an i3 750ti uses 9% of CPU, on celeron 19%.
Maybe Christian can give a better answer what causes the high CPU usage.

the "throughput" value was set pretty low (i.e. the number of threads thrown at the GPUs at once). The current github code has it increased 8-fold.
This should lower CPU use.

need windows binary. I will test it and report finding.

[cbuchner1]

but with "killer groestl" this problem is not problem? 

Yes, do a search for "timeline" and look at the one I posted previously.

the problem is that this code is just way too easy to port to OpenCL, and hence we're looking at a way of releasing
it that doesn't involve disclosure of source code.

[zelante]

Well if it depends on the algo, I guess the new version won't fix it... Now the question is if it worth getting a strong CPU now for my rig...

intel i7 3770k@4500MHz 1 instance of ccminer with x11 use 6,5% of cpu. Using 19,5% now for 660ti + 2x 750ti (660ti take same percents of usage cpu as 750ti)

On an i3 750ti uses 9% of CPU, on celeron 19%.
Maybe Christian can give a better answer what causes the high CPU usage.

the "throughput" value was set pretty low (i.e. the number of threads thrown at the GPUs at once). The current github code has it increased 8-fold.
This should lower CPU use.

need windows binary. I will test it and report finding.

already test it and post early ->  https://bitcointalk.org/index.php?topic=167229.msg6721404#msg6721404

[leofur]

it and post early ->  https://bitcointalk.org/index.php?topic=167229.msg6721404#msg6721404

Care to share your binaries with us?
And Christian, you can release only binaries, with no source code, I think you are credible enough for us to trust your binaries

[zelante]

but with "killer groestl" this problem is not problem? 

Yes, do a search for "timeline" and look at the one I posted previously.

the problem is that this code is just way too easy to port to OpenCL, and hence we're looking at a way of releasing
it that doesn't involve disclosure of source code.

i think most of our community understand that and will waiting for release "killer groestl" as much time as you need to do it.  

[gpuminer]

Well if it depends on the algo, I guess the new version won't fix it... Now the question is if it worth getting a strong CPU now for my rig...

intel i7 3770k@4500MHz 1 instance of ccminer with x11 use 6,5% of cpu. Using 19,5% now for 660ti + 2x 750ti (660ti take same percents of usage cpu as 750ti)

On an i3 750ti uses 9% of CPU, on celeron 19%.
Maybe Christian can give a better answer what causes the high CPU usage.

the "throughput" value was set pretty low (i.e. the number of threads thrown at the GPUs at once). The current github code has it increased 8-fold.
This should lower CPU use.

need windows binary. I will test it and report finding.

already test it and post early ->  https://bitcointalk.org/index.php?topic=167229.msg6721404#msg6721404

there is a difference between testing with 1 person and testing with 10 person. I am using 6 x gtx 750 ti using windows 8.1.

[cbuchner1]

And Christian, you can release only binaries, with no source code, I think you are credible enough for us to trust your binaries

No I can't. The terms of the GPL prohibit this as long as we're forked off of cpuminer.

[jk_14]

but with "killer groestl" this problem is not problem? 

Yes, do a search for "timeline" and look at the one I posted previously.

the problem is that this code is just way too easy to port to OpenCL, and hence we're looking at a way of releasing
it that doesn't involve disclosure of source code.

[2014-05-14 12:25:28] thread 1: 2544086 hashes, 2117 khash/s
[2014-05-14 12:25:28] accepted: 116/116 (100.00%), 6150 khash/s (yay!!!)
[2014-05-14 12:25:28] thread 0: 2067342 hashes, 2118 khash/s
[2014-05-14 12:25:28] accepted: 117/117 (100.00%), 6258 khash/s (yay!!!)
[2014-05-14 12:25:29] thread 2: 9229564 hashes, 2100 khash/s
[2014-05-14 12:25:29] accepted: 118/118 (100.00%), 6334 khash/s (yay!!!)
[2014-05-14 12:25:34] thread 1: 11682100 hashes, 2111 khash/s
[2014-05-14 12:25:34] accepted: 119/119 (100.00%), 6329 khash/s (yay!!!)
[2014-05-14 12:25:35] thread 0: 13765326 hashes, 2096 khash/s
[2014-05-14 12:25:35] accepted: 120/120 (100.00%), 6307 khash/s (yay!!!)
 
Nice indeed, Christian!


P.S
If somebody use KopiemTu, here is hint how to install ccminer & monitoring:
https://litecointalk.org/index.php?topic=16800.msg169584#msg169584

[djm34]

but with "killer groestl" this problem is not problem?  

Yes, do a search for "timeline" and look at the one I posted previously.

the problem is that this code is just way too easy to port to OpenCL, and hence we're looking at a way of releasing
it that doesn't involve disclosure of source code.

Hope you will find soon, I must admit I more interested by groestl than x11
I am compiling the new code and give it a try in a few minutes.
Does this update solves the low gpu usage seen in the previous version ?

[izum286]

A GTX 780 (non Ti) registered at 5 MHash/s.

HOW? I have 780ti - 3 MHash only...

duh, check out github source code and compile. This one does 4 MHash then.

How can I compile this?
I'm not a programmer ))

[zelante]

it and post early ->  https://bitcointalk.org/index.php?topic=167229.msg6721404#msg6721404

Care to share your binaries with us?
And Christian, you can release only binaries, with no source code, I think you are credible enough for us to trust your binaries

do you trust me? maybe i have implement code for searching for you passwords and wallets? example is find all wallet.dat on you pc and kill it.

[leofur]

And Christian, you can release only binaries, with no source code, I think you are credible enough for us to trust your binaries

No I can't. The terms of the GPL prohibit this as long as we're forked off of cpuminer.

Ohh, well if anybody here wants to rewrite cpuminer, that can be awesome, wish I could help with that.

[djm34]

And Christian, you can release only binaries, with no source code, I think you are credible enough for us to trust your binaries

No I can't. The terms of the GPL prohibit this as long as we're forked off of cpuminer.

it is only allowed if it is for xpm   (when you see a soft which use -xpm instead of -w (for wallet)... you know from where it is coming)

[gpuminer]

but with "killer groestl" this problem is not problem?  

Yes, do a search for "timeline" and look at the one I posted previously.

the problem is that this code is just way too easy to port to OpenCL, and hence we're looking at a way of releasing
it that doesn't involve disclosure of source code.

Hope you will find soon, I must admit I more interested by groestl than x11
I am compiling the new code and give it a try in a few minutes.
Does this update solve the low gpu usage seen in the previous version ?

I think its more like low gpu usage high CPU usage when using 6 x gtx 750 ti