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

[bziubek]

Got it running on Nvidia Quadro FX 1800 at ~11kh/s on default settings.

[crazy_rabbit]

Does this make Nvidia cards realistic for mining? OR are they still underperforming comparatively?

[Luckybit]

I'm new to mining. I have a Nvidia Gforce 9400 GT. Is it worth it to try mining or should I just wait until I can afford to buy a new specialized setup?

[salfter]

At the low end of the performance spectrum, the 9500GT in my work computer's getting about 7.5 kH/s. It leaves my system fairly unresponsive, though.  With something like cgminer, you could back off the intensity to get acceptable desktop performance on a system that's not dedicated to mining.  Is there a similar option here?  I didn't see anything in cudaminer -h or in a quick perusal of the source code.  I have a config file set up, but it currently only has algo, url, and user options defined.

I'm not expecting any miracles performance-wise, but if I could squeeze out a few more coins without impacting this computer's normal functions, that would be nice.

[cbuchner1]

Hi

I am still working on this tool.

I am now trying to allocate the card's VRAM in chunks, as the WDDM driver model of Windows Vista, 7 and 8 enforces some upper limit on single memory allocations. Also I will be trying to solve the crashes that some of you were seeing with cards like the GTX 570 or 580 where the launch configuration would require more memory than I was able to allocate.

I am working on a better autotune features that tries a lot of kernel launch configurations for best speed. This has already brought my GTX 460 up from 56 to 62 kHash. We will also have override options for any automatic card setting.

I am also investigating why some GT2xx and 8/9 series cards may initially report a higher kHash rate than is later achieved - e.g. GTX 260 reports up to 55 kHash initially but later achieves only 42.

Christian

[TheBigYak]

How much VRAM is on your Card? What model is it? Do you have Drivers above 304.something? (required for CUDA 5.x)

Christian

Card is a GTX 560 Superclocked, herehttp://www.amazon.com/gp/product/B0050I1PGU/ref=oh_details_o01_s03_i00?ie=UTF8&psc=1.

1024MB of RAM, and yes, using the 306.97 driver stack.  Windows 8 is the OS.

[cbuchner1]

I just posted an update (April 6th), which adds an autotuning feature, so it maximizes performance a bit more. It will also never try to use more memory than we can allocate on the card, which should fix the crashes on startup that some people were seeing.



Remind me to update the version shown in the console, next time I publish a release.

Due to autotuning, my GTX 460 now does 64 kHash/s. The previous version only managed 56 kHash/s
The GTX 260 in above example picks a launch configuration of 54x3 instead of the default 27x3.

When autotune indicates a chosen launch configuration that is prefixed with "S" it means that this uses the spinlock CUDA kernel that works best on some Fermi cards (GT 4xx and 5xx) and on Kepler cards (GT 6xx). Future releases will have a command line switch to override the launch configuration per card (which automatically disables autotuning for this card).

[cbuchner1]

Reverted back to 04-04-2013 version for now.

there is a --no-autotune switch if this feature doesn't do much for you, or if you need to restart your miner once every few minutes ( other than for configuration set up, I don't see any need for so many restarts ).

Also look at the above screenshot to see how it is expected to be working. I guess in your case the Autotune simply wasn't done yet when you aborted. Generally, the more memory is on your card the more combinations autotune will attempt. The LONGPOLL detection will occasionally output messages even during autotune.

Christian

[peacefulmind]

I have 2 TITANS I could try this on.  In the past I got about 350k/H for both cards together using cgminer --script

[cbuchner1]

I haven't optimized for Titan yet. I expect a single card could do 300kHash/sec when using the new Funnel Shifter hardware feature.

I hope you bought them for gaming, not for mining

[peacefulmind]

I haven't optimized for Titan yet. I expect a single card could do 300kHash/sec when using the new Funnel Shifter hardware feature.

I hope you bought them for gaming, not for mining

They are for gaming.  They are great for driving new games in 5760x1200 surround. 

Still might use them when not gaming sometimes if I got a combined 700k/H out of both to just run while I sleep.  Also I just enjoy the experimentation.

[Btuser1]

cbuchner1: It works fine with my nvidia card - 50Kh/s 
But WHY IT DOESN'T MINE SHARES on pool?  Console window shows "accepted" messages but my pool shows that your miner doesn't mine any shares(worker has 0 kh/s, while cpuminer's worker has 10kh/s)

When you planning to release real working(shares mining) version of cudaminer?

[blastbob]

Ill give it a spin tomorrow on my dusty geforce 580 gtx

[MrLetter]

EVGA GTX 570, getting just shy of 140 KH/s on each solve. Initial auto config gave me something around 170KH/s when I started it up.

GeForce driver: 314.22
EVGA Precision settings: GPU 910MHz; Memory 2100 MHz; 1025mv (If you guys have suggestions for changes to these settings, I'd love to hear them.)

--> EDIT: I almost forgot, my temp stayed between 69c and 70c which is 4c to 6c cooler than reaper, and 5c to 8c cooler than btc mining. Which is nice. <--

-------------

First of all thank you for working on this, I've been playing around with Reaper for the last few weeks attempting to find the sweet spot in the config. The best I got was just shy of 90KH/s. With that said, it took me a hot minute to figure out how to set this up since the readme file didn't say anything. lol. My suggestion would be to add some info about that, and other functions for those of us switching to this from reaper, or other non CGminer type miners. Along with a little background about the other required programs. My only real request at this moment would be a graceful shutdown feature like what reaper has, where it doesn't crash the driver.

My observations were that I really couldn't use anything else while this was running. Firefox took minutes to respond, often switching to "not responding" or otherwise hanging. So my habit of watching Netflix while letting reaper and scryptminer-gui run is now broken. Speaking of scyptminer my blond moment was not turning it off during my initial run of this program (the numbers above are from the second run). The only oddity was that print-screen didn't capture anything.

Other system info:
i7-2600K @ 3.8
16GB RAM
Win7 64-bit

[wndrbr3d]

Seeing odd performance metrics on my 690.

Both GPU's sit around 25-35% utilization and it's not reporting any hash rates, it just starts up, says it's running two threads and that's it.

Not sure if it's an issue specific to the 690 or an SLI setup. Curious if anyone else is having similar issues.

[cbuchner1]

Hi all,

today I've been refactoring the CUDA kernels. I was able to cut shared memory use in half and also lowered the register use. This leads to higher occupancy on the multiprocessor units, resulting in higher speeds !

GTX 460 now solving 80 kHash/s (previously around 62-64). I will give it a spin on more nVidia devices I own, and within a short time frame you can expect an update. EDIT: With the same optimization the GTX 260 performance goes down from 42 to 31 kHash. Meh.

MrLetter: you will soon get even more kHash out of the GTX 570. I believe the CPU is stealing those 30 kHash/sec because it has to do the HMAC 256 part of scrypt. And once the CPU core is 100% pegged, it can't provide the GPU with work units fast enough. I will have to include the SSE2 optimized assembly code from cpuminer to fix this.

wndrbr3d: pass the --no-autotune option please or wait a few minutes more... It could also be stuck at requesting a LONGPOLL connection with the pool. Longpoll can be disabled in options, too.

Some of you really want to use their desktops while mining. We can have that, but it will slow down your kHash/sec of course...

Christian

[aigeezer]

Thanks for the CUDA miner. I've certainly got uses for it.

I gave it a quick try hoping to use the coinotron pool and got:

E:\cudaminer-2013-04-06>cudaminer -t 1 -o http://coinotron.com:3334 -O myworkername:mypassword
           *** CudaMiner for nVidia GPUs by Christian Buchner ***
                     This is version 2013-04-04 (alpha)
        based on pooler-cpuminer 2.2.3 (c) 2010 Jeff Garzik, 2012 pooler
               Cuda additions Copyright 2013 Christian Buchner
           My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-04-07 21:36:47] 1 miner threads started, using 'scrypt' algorithm.
[2013-04-07 21:36:47] HTTP request failed: Empty reply from server
[2013-04-07 21:36:47] json_rpc_call failed, retry after 15 seconds
[2013-04-07 21:37:02] HTTP request failed: Recv failure: Connection was reset
[2013-04-07 21:37:02] json_rpc_call failed, retry after 15 seconds
^C
E:\cudaminer-2013-04-06>

I'm a tired old man - see anything wrong with what I did? I used actual name and password, of course,
but I'll double-check them. 

[wndrbr3d]

wndrbr3d: pass the --no-autotune option please or wait a few minutes more... It could also be stuck at requesting a LONGPOLL connection with the pool. Longpoll can be disabled in options, too.

Still no bueno, although you're getting closer This version doesn't crash on start up, but now it just seems to not mine. Not entirely sure yet.

Here's what I see happening in GPU-Z:



GPU load seems to go up and down some while the host CPU (i5-3570) gets pegged. Perhaps an issue with the CPU not being able to feed the massive amounts of SP's with the SHA256 calc happening on the CPU?

[dragon2nd]

Here's my new pet project. I started this during the easter holidays, and this uses CUDA to squeeze up to 100%  more performance out of nVidia cards - as compared to OpenCL mining applications.  

Please carefully inspect the README.txt file before use. Usage is pretty much identical to pooler's cpuminer

cudaminer-2013-04-06.zip (952 KB)
https://mega.co.nz/#!pUIgkA5Q!Tzy1CEUdyfCOOsyRyueW4QrH6Kx-OX-loLxot_4-dcI

Let me know if this works or fails for you. Post your GPU specs and achieved kHash/sec please. Also let me know if you experience any crashes or if you have specific suggestions for improvement.

If you find this useful, please donate a little. How about the first litecoin you successfully mine with this app? LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

I might be improving this a bit more during the upcoming weeks. Porting this back to Linux also shouldn't be too hard, really.

Required dependencies for building on Windows:
pthreads: http://sourceware.org/pthreads-win32/
OpenSSL-Win32: http://slproweb.com/download/Win32OpenSSL-1_0_1e.exe
curl-7.29.0: http://curl.haxx.se/download/curl-7.29.0.zip
Better build for Win32, as there is no benefit compiling a 64bit binary.

Christian

Can anyone port it to Linux, please?

[wndrbr3d]

wndrbr3d: pass the --no-autotune option please or wait a few minutes more... It could also be stuck at requesting a LONGPOLL connection with the pool. Longpoll can be disabled in options, too.

Still no bueno, although you're getting closer This version doesn't crash on start up, but now it just seems to not mine. Not entirely sure yet.

Here's what I see happening in GPU-Z:



GPU load seems to go up and down some while the host CPU (i5-3570) gets pegged. Perhaps an issue with the CPU not being able to feed the massive amounts of SP's with the SHA256 calc happening on the CPU?

Dove into the code and I think there's something going on with the chunked memory allocation routine in find_optimal_blockcount(). I had it log out the warps through the loop, you can see the two threads doing their allocations but things start getting REAAAALLY slow...



Not sure what could cause this. When I set -t 1, it flies through this no problem so I imagine it has something to do with the two GPU threads being kicked off at once.

Hope this helps!