python OpenCL bitcoin miner

[fisuk]

Finally got my hands on a GTX 275. Reporting stats:

CPU: Core2 Extreme QX6700, 2.67 GHz
OS: Gentoo Linux 2.6.34
Client: Classic 0.3.19 (local build)
khash/sec: ~5000

GPU: Nvidia GeForce GTX 275
Miner: m0mchil's OpenCL (2010-12-20)
Options: -r 3 -f 10 -w 256
khash/sec: ~49500

I have an old motherboard (PCIe 1.0a, etc.) which has a lot of strain on it, so your results with the same CPU or GPU could be better. It seems I should've gone for an ATI radeon though.

Yeah, I guess the problem lies with Nvidia's OpenCL drivers which are less than optimal. If there was a CUDA miner available, Nvidia users would probably get more comparable results.. So, who's up to the task?

[brocktice]

Finally got my hands on a GTX 275. Reporting stats:

CPU: Core2 Extreme QX6700, 2.67 GHz
OS: Gentoo Linux 2.6.34
Client: Classic 0.3.19 (local build)
khash/sec: ~5000

GPU: Nvidia GeForce GTX 275
Miner: m0mchil's OpenCL (2010-12-20)
Options: -r 3 -f 10 -w 256
khash/sec: ~49500

I have an old motherboard (PCIe 1.0a, etc.) which has a lot of strain on it, so your results with the same CPU or GPU could be better. It seems I should've gone for an ATI radeon though.

Yeah, I guess the problem lies with Nvidia's OpenCL drivers which are less than optimal. If there was a CUDA miner available, Nvidia users would probably get more comparable results.. So, who's up to the task?

Nvidia hardware sucks for mining at the hardware level. CUDA won't help.

[GeorgeH]

Does anyone know what the optimal settings are for a GTX 580?

Thanks,
DiSTANT

Sell it and buy an ATI card.

[jiffy]

Hi,

poclbm finally told me that a block was accepted (after 3 days with 0,23Ghash/s).

bitcoind getinfo shows no balance of +50 until now, though.

I read somewhere that this might take very long because the new block is checked very intensively. Couldn't find the text again, though.

So can you tell me that I can calm down or went something wrong and the 3 days of work are gone for somebody else?

[davout]

you have to wait 120 blocks before the coins mature,
otherwise you might spend coins that were generated on a chain fork

[jiffy]

you have to wait 120 blocks before the coins mature,
otherwise you might spend coins that were generated on a chain fork

Thanks a lot for quick reply, davout!

Is it possible to see this on blockexplorer.com? What do I have to look for?

[theymos]

Is it possible to see this on blockexplorer.com? What do I have to look for?

You can see the block on BBE as soon as it is generated (1-minute delay). If it shows up there, it probably won't be reversed, though from time to time this does happen. It won't appear in your balance until it is 120 blocks deep.

The block hash is printed to debug.log by Bitcoin when you find a block. poclbm itself might also do this.

Code:

proof-of-work found  
  hash: 000000001cd845dc081bd5c2f0165f6101edee254a0f716701b70c1522758b69

You can use the hash to search BBE for the block.

[gohan]

Yeah, I guess the problem lies with Nvidia's OpenCL drivers which are less than optimal. If there was a CUDA miner available, Nvidia users would probably get more comparable results.. So, who's up to the task?

Nvidia hardware sucks for mining at the hardware level. CUDA won't help.

Hi, are there any references you can recommend about this (which thread to read about Nvidia's deficiencies, etc.)? I'm interested in the implementations but I won't get in too deep if there isn't much to be done on CUDA.

[GeorgeH]

I don't have any references on hand, but here's the Reader's Digest version: nVidia GPUs are built to do fast floating point math. Cryptological algorithms like Bitcoin's are computed using integer math. On the flip side, ATI's GPUs have strong integer math capabilities. ATI cards generally have 4-6x the performance of comparably priced nVidia cards.

[brocktice]

Yeah, I guess the problem lies with Nvidia's OpenCL drivers which are less than optimal. If there was a CUDA miner available, Nvidia users would probably get more comparable results.. So, who's up to the task?

Nvidia hardware sucks for mining at the hardware level. CUDA won't help.

Hi, are there any references you can recommend about this (which thread to read about Nvidia's deficiencies, etc.)? I'm interested in the implementations but I won't get in too deep if there isn't much to be done on CUDA.

Nvidia and ATI take different performance approaches, and even the newly-announced AMD/ATI GPUs take a slightly different approach than the previous generation. (Approach meaning number of ALUs, memory bandwidth, etc etc.) What this means is that ATI and Nvidia are lately neck-and-neck when it comes to average gaming performance, but for number crunching, especially integer number crunching, ATI/AMD 5xxx series currently hold the worldwide performance, price, and performance/price and performance/watt top spot, by a long shot.

[gohan]

Ah, Integers! Thank you Smartzkid & brocktice. I thought it would be something more subtle (and bound to change), that's why I went for nVidia. Better warn suckers like me not to buy nVidia for bitcoin mining.

[brocktice]

Ah, Integers! Thank you Smartzkid & brocktice. I thought it would be something more subtle (and bound to change), that's why I went for nVidia. Better warn suckers like me not to buy nVidia for bitcoin mining.

It's more than that, basically it's number of stream processors. A 5970 still wins vs. any nvidia for floating point math, even more so when you factor in power and price.  The major downside with ATI is that their GPGPU toolchain is less mature than that of Nvidia. A cluster provider my company uses refuses to put ATI cards in their cluster for that reason. They don't trust them. ATI/AMD cards also not currently designed for headless operation, unlike Tesla cards.

[m0mchil]

Updated to support latest change of slush's pool which now returns JSON RPC error when its back-end server is down. Original bitcoind never returns such error as a response to getwork(data).

Summarized, this should help miner continue after pool downtime.

[slush]

Summarized, this should help miner continue after pool downtime.

Thank you for this update. I believe it will improve overall pool stability. I had to revert json error back to original HTML output, because current miners are crashing. But I will change it to json response after few days, when majority of users will use fixed miners.

[Eloar]

I have some problems with that. First I can't download binary package. Could someone send it to somekind mirror?

I'm trying to get it working from script, but it is not so easy. I can't make pyopencl installation right. My siteconf.py is:

BOOST_INC_DIR = [r'C:\Program Files (x86)\boost\boost_1_44']
BOOST_LIB_DIR = [r'C:\Program Files (x86)\boost\boost_1_44\lib']
BOOST_COMPILER = 'msvc'
BOOST_PYTHON_LIBNAME = ['boost_python-vc90-mt-1_44']
USE_SHIPPED_BOOST = False
CL_TRACE = False
CL_ENABLE_GL = False
CL_INC_DIR = ['C:\Program Files (x86)\ATI Stream\include']
CL_LIB_DIR = ['C:\Program Files (x86)\ATI Stream\lib\\x86']
CL_LIBNAME = ['OpenCL']
CXXFLAGS = ['/EHsc', '/DBOOST_PYTHON_NO_PY_SIGNATURES']
LDFLAGS = ['/FORCE']

I get as output from installation (I don't see any errors :/ ) :

running install
install_dir C:\Program Files (x86)\python\Lib\site-packages\
running bdist_egg
running egg_info
writing requirements to pyopencl.egg-info\requires.txt
writing pyopencl.egg-info\PKG-INFO
writing top-level names to pyopencl.egg-info\top_level.txt
writing dependency_links to pyopencl.egg-info\dependency_links.txt
writing requirements to pyopencl.egg-info\requires.txt
writing pyopencl.egg-info\PKG-INFO
writing top-level names to pyopencl.egg-info\top_level.txt
writing dependency_links to pyopencl.egg-info\dependency_links.txt
reading manifest file 'pyopencl.egg-info\SOURCES.txt'
reading manifest template 'MANIFEST.in'
writing manifest file 'pyopencl.egg-info\SOURCES.txt'
installing library code to build\bdist.win32\egg
running install_lib
running build_py
running build_ext
building '_cl' extension
error: None

but when I try to run poclbm.py I get:

Traceback (most recent call last):
  File "poclbm.py", line 3, in <module>
    import pyopencl as cl
ImportError: No module named pyopencl

I was trying for few hours lot of configurations and nothing worked. Pynum, scipy and so are installed and works. All is for x86 installed on Win7 Pro x86_64 on ATi Radeon HD4870. New ATI Stream SDK and CCC are installed, and I tested some apps from examplest that are working.

[chris200x9]

A couple questions, how do I confirm this is running for me? In the bitcoin app I don't see and khashes/s I only see it in the terminal in which I ran this from. Second question, my cpu is not being used at all can I use my cpu with the regular bitcoin client to work in parellel. Lastly can I use another computer to work in paralell?

[Mahkul]

how do I confirm this is running for me? In the bitcoin app I don't see and khashes/s I only see it in the terminal in which I ran this from.

If you can see the khash/s rate this means your miner is working. Otherwise it would give you an error message, like: Unable to connect to Bitcoin RPC or Bitcoin isn't running.

Second question, my cpu is not being used at all can I use my cpu with the regular bitcoin client to work in parellel.

Yes, just enable "Generate coins".

Lastly can I use another computer to work in paralell?

Yes, as long as you allow its IP in the Bitcoin config file using the rpcallowip option.

Please note that all these questions have already been answered on the forum. Next time try to use Search.

[chris200x9]

ok Mahkul sorry and thanks so much!

[FairUser]

Edit: False alarm, please ignore.

Hi,

I'm seeing several repeat "accepted" hashes.  Here's an example.

01/01/2011 18:09, e62d8177, accepted
01/01/2011 18:09, e62d8177, accepted
01/01/2011 18:09, a753edb1, accepted
01/01/2011 18:09, c46b1614, accepted
01/01/2011 18:10, 51ccbc9c, accepted
01/01/2011 18:10, 51ccbc9c, accepted
01/01/2011 18:11, 2b411e8b, accepted
01/01/2011 18:11, 2b411e8b, accepted
01/01/2011 18:11, 2b411e8b, accepted
01/01/2011 18:13, 3ddfdb84, accepted
01/01/2011 18:13, 9e29c06c, accepted
01/01/2011 18:13, bdc65504, accepted
01/01/2011 18:13, bdc65504, accepted
01/01/2011 18:14, f4d73c8f, accepted
01/01/2011 18:14, 7229e2dc, accepted
01/01/2011 18:15, af861f38, accepted
01/01/2011 18:15, c5d3b074, accepted
01/01/2011 18:16, c446c905, accepted
01/01/2011 18:16, 5e9072e6, accepted
01/01/2011 18:16, 5e9072e6, accepted

It seems to be getting more frequent over the last few days too, as in I'm now see 3 to 4 repeat "accepted" hashes.
The number of "shares" I get from bitcoin.cz doesn't increase with repeats either.

What would cause the same hash being accepted more than once?
Is my client computing the same hash more than once?  If so, why?

Thanks for the great code.  Getting ~150,000 khash/s.....I think.

[Edogaa]

alright, I followed the instructions on that site... but why do I keep getting this error regardless of whether I open the .bat file or do so manually?


C:\bitcoin\poclbm_py2exe_20101228\poclbm.exe --user=no --pass=no
Traceback (most recent call last):
  File "poclbm.py", line 3, in <module>
  File "pyopencl\__init__.pyc", line 3, in <module>
  File "pyopencl\_cl.pyc", line 12, in <module>
  File "pyopencl\_cl.pyc", line 10, in __load
ImportError: DLL load failed: The specified procedure could not be found.



I thought I was just doing something wrong at first, but even when I followed the instruction on the blog, word for word, with a slight variation in version numbering I get this error. I don't get what this means, and the poclbm.py has 0 lines in it, and takes up 0 KB. :/