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

[cbuchner1]

It seems that I somehow got lost in my CUDA optimization. After 3 days of trying, I ended up having less kHash/sec compared to the April 22nd version.

So I re-start my optimization endeavour from scratch. This time I throw out the old S kernels, and my optimized kernels will bear the "S" prefix instead. Now during autotune I can directly compare "S" vs. non-S performance, and if there are any performance benefits or penalties, I can immediately quantify them.

I am at a point where seemingly harmless changes make the compiler do something different, and it ends up having an unexpected performance penalty.

Christian

[1714241180]

1714241180

[1714241180]

1714241180

[1714241180]

1714241180

[Cheshyr]

It seems that I somehow got lost in my CUDA optimization. After 3 days of trying, I ended up having less kHash/sec compared to the April 22nd version.

So I re-start my optimization endeavour from scratch. This time I throw out the old S kernels, and my optimized kernels will bear the "S" prefix instead. Now during autotune I can directly compare "S" vs. non-S performance, and if there are any performance benefits or penalties, I can immediately quantify them.

I am at a point where seemingly harmless changes make the compiler do something different, and it ends up having an unexpected performance penalty.

Christian

Solar flare.  Definitely the solar flares fault.

At least you've got version control.

[nst6563]

damn those solar flares.  piss me off they do.

[coalescent]

https://i.imgur.com/dTlL0ZI.png

Looks like the 320.00 beta's improved my performance quite a bit. On the 4-22 build I was getting about 205kH/s (125/400 boost). But now am getting upwards of 220kH/s. Pretty decent.

[termhn]

Any chance of a version of cudaminer for this when it comes out? Been watching it closely and it looks really cool. https://bitcointalk.org/index.php?topic=169204.0

[FalconFour]

Solar flare.  Definitely the solar flares fault.

At least you've got version control.

* pictures cbuchner1's reply...

Yeah, ah... version control.
*downloads the archive file from the first post here*
*extract, overwrite*
Yep, version control. Good thing for that.

[Minas3d]

I've been trying to use this miner since the first release on my 9800GTX, but everytime I start it up it crashes the second it starts mining. I've tried every and all settings on different releases and it continues to crash, I have no idea what's causing it.

[cbuchner1]

At least you've got version control.

*downloads the archive file from the first post here*
*extract, overwrite*
Yep, version control. Good thing for that.

Exactly. While we use SVN at work, at home all I've got is backup copies of directories and files. No git, no svn, no cvs, no mercurial.

[cbuchner1]

I've been trying to use this miner since the first release on my 9800GTX, but everytime I start it up it crashes the second it starts mining. I've tried every and all settings on different releases and it continues to crash, I have no idea what's causing it.

Are you using nVidia driver is release 304.xxx or later? This reminds me that the next version should really output verbose error messages. I've had a couple of reports where the kernels would just not work for some reason, yet the CUDA card was identified correctly with name and compute capability.

[cbuchner1]

Hmm, CPU usage decreases significantly on Windows when I split the CPU workload (SHA-256 hashing) into exactly N equal chunks where N is the number of CPU cores on the system.

I think previously I did something wrong in splitting up the CPU workload. So it actually had to do work harder than it needed to. Hmm.. Bad bug: A significant part of the work was overlapping and therefore redundant. Also, no one ever took a closer look at my source code.  

CPU usage now hovers around 15% when I produce 350-400kHash/sec on an AMD Phenom II 1055T processor.

Christian

[beppe]

hi, i'm using a gt-430 but can't get anything to work.

when I launch
cudaminer.exe -o http://newlc.ozco.in:9332 -u xxxxxxxxxxx -p xxxxxxxxx -d 0
i get

Code:

           *** CudaMiner for nVidia GPUs by Christian Buchner ***
                     This is version 2013-04-22 (alpha)
        based on pooler-cpuminer 2.2.3 (c) 2010 Jeff Garzik, 2012 pooler
               Cuda additions Copyright 2013 Christian Buchner
           My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-04-26 20:10:35] 1 miner threads started, using 'scrypt' algorithm.
[2013-04-26 20:10:35] Long-polling activated for http://newlc.ozco.in:9332/LP
[2013-04-26 20:10:35] GPU #0: starting up...

[2013-04-26 20:10:35] GPU #0:  with compute capability 716.25
[2013-04-26 20:10:35] GPU #0: interactive: 0, tex-cache: 0 , single-alloc: 0
[2013-04-26 20:10:35] GPU #0: Performing auto-tuning (Patience...)
[2013-04-26 20:10:35] GPU #0:    0.00 khash/s with configuration  0x0
[2013-04-26 20:10:35] GPU #0: using launch configuration  0x0

then it crashes and I see the window "windows is searching for a solution"

if I launch cudaminer with --no-autotune it doesn't crash, but all I get is

[2013-04-26 20:14:21] GPU #0: , 0 hashes, 0.00 khash/s

what can I do?

[cbuchner1]

[2013-04-26 20:10:35] GPU #0:  with compute capability 716.25

See, it can't even open your GPU. No name indicated and compute capability is bogus.

Try updating your driver. It must be new enough to support CUDA 5.0.

[dbabo]

.. Also, no one ever took a closer look at my source code.  

not exactly. Buddy of mine hovered over it trying to get it to work on 4.2.
I'm not sure what came out of that yet.

..
CPU usage now hovers around 15% when I produce 350-400kHash/sec on an AMD Phenom II 1055T processor.

Christian

does it mean there is a newer version i can try ?

[cbuchner1]

Nope, not yet.

[Schleicher]

See, it can't even open your GPU. No name indicated and compute capability is bogus.

Maybe some code like this would be helpful?

Code:

error=cudaGetDeviceCount(&deviceCount);
if(error!=cudaSuccess)
{
	if(error==cudaErrorNoDevice)
		cout << "No CUDA capable device found" << endl;
	else
		cout << cudaGetErrorString(error) << endl;
}

[cbuchner1]

I think this is the next item on my TODO list.

[BrewCrewFan]

hi, i'm using a gt-430 but can't get anything to work.

when I launch
cudaminer.exe -o http://newlc.ozco.in:9332 -u xxxxxxxxxxx -p xxxxxxxxx -d 0
i get

Code:

           *** CudaMiner for nVidia GPUs by Christian Buchner ***
                     This is version 2013-04-22 (alpha)
        based on pooler-cpuminer 2.2.3 (c) 2010 Jeff Garzik, 2012 pooler
               Cuda additions Copyright 2013 Christian Buchner
           My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-04-26 20:10:35] 1 miner threads started, using 'scrypt' algorithm.
[2013-04-26 20:10:35] Long-polling activated for http://newlc.ozco.in:9332/LP
[2013-04-26 20:10:35] GPU #0: starting up...

[2013-04-26 20:10:35] GPU #0:  with compute capability 716.25
[2013-04-26 20:10:35] GPU #0: interactive: 0, tex-cache: 0 , single-alloc: 0
[2013-04-26 20:10:35] GPU #0: Performing auto-tuning (Patience...)
[2013-04-26 20:10:35] GPU #0:    0.00 khash/s with configuration  0x0
[2013-04-26 20:10:35] GPU #0: using launch configuration  0x0

then it crashes and I see the window "windows is searching for a solution"

if I launch cudaminer with --no-autotune it doesn't crash, but all I get is

[2013-04-26 20:14:21] GPU #0: , 0 hashes, 0.00 khash/s

what can I do?

Stupid question ( Heck I just now figured out how to get this thing to launch myself ) .... but did you make a launch.bat file? When I first tried to use cuda, I got something kinda like that... then I made the launch.bat, put the info needed in it :

cudaminer -o http://website:port -O"username:password" -i 1 ( take out the " of course... leaving a space)

then I just double click on it and it launches no probs for me. You can add flags to this as needed ( like remove autotune and add you own *x* to it ectect)

I wish the directions were more clear than "dl this and that and run"...makes it hard for those of us that are not very good at stuff like this.

Yeah I see what you did there... thats not right. Make the launch.bat file with your notepad...make sure when you save it , you save it as "launch.bat" without the ".... chose "all files" when you save it so it will save as a .bat file instead of a notepad ...and in the folder you have the cuda.exe in. Then all you should have to do is double click on the .bat file you just made and it should launch.

[beppe]

[2013-04-26 20:10:35] GPU #0:  with compute capability 716.25

See, it can't even open your GPU. No name indicated and compute capability is bogus.

Try updating your driver. It must be new enough to support CUDA 5.0.

thank you and to all the others, I solved installing latest 314.22 nvidia driver!

now I see

Code:

[2013-04-26 22:10:48] GPU #0: GeForce GT 430, 81408 hashes, 18.13 khash/s

same speed of cgminer but it's ok!

[Lacan82]

[2013-04-26 20:10:35] GPU #0:  with compute capability 716.25

See, it can't even open your GPU. No name indicated and compute capability is bogus.

Try updating your driver. It must be new enough to support CUDA 5.0.

thank you and to all the others, I solved installing latest 314.22 nvidia driver!

now I see

Code:

[2013-04-26 22:10:48] GPU #0: GeForce GT 430, 81408 hashes, 18.13 khash/s

same speed of cgminer but it's ok!

Try it with -i 0 at the end of the launch

[BrewCrewFan]

[2013-04-26 20:10:35] GPU #0:  with compute capability 716.25

See, it can't even open your GPU. No name indicated and compute capability is bogus.

Try updating your driver. It must be new enough to support CUDA 5.0.

thank you and to all the others, I solved installing latest 314.22 nvidia driver!

now I see

Code:

[2013-04-26 22:10:48] GPU #0: GeForce GT 430, 81408 hashes, 18.13 khash/s

same speed of cgminer but it's ok!

Try it with -I 0 at the end of the launch

that is a -i and not an uppercase.

He might also try using the -l flag also to see how it goes. My 450 has one of -l 40x2 I think.