python OpenCL bitcoin miner

[Compute4Cash]

Compute4Cash is no more a scam than Bitcoin, and Bitcoin is certainly not a scam either.  Please refrain from slandering the Compute4Cash name - it really isn't necessary nor is it called for.

While of course we are not happy with the the orchestrated efforts to 'expose' Compute4Cash, we knew it was only a matter of time before this happened and we are not mad about it.  What does bother us however are the comments that essentially claim that Compute4Cash is screwing people over.  Yes, there are other alternatives to Compute4Cash that should yield higher profits for users, but Compute4Cash offers additional value over what the alternatives provide, and there is of course a premium to be paid for that.  Some may believe that this 'premium' is too high right now, but that only applies for right now.  Things are constantly changing and we expect to adjust our premium to be more reasonable in the near future, and we also expect that as the Bitcoin system and markets continue to grow and adjust themselves that this will also bring our premium to a more reasonable level.

Another thing that bothers us is the attempt to persuade our users to jump ship - one thing that Compute4Cash has done very differently from other alternatives is that we have engaged in paid-for advertising to grow our user base.  We have paid a substantial amount of money to share this opportunity with other people, knowing also that at some point in the future this would equate to advertising for Bitcoin and getting more people involved.  We therefore have invested in each and every user we have, so in a sense when you tell people to leave Compute4Cash you are running off with a portion of our investment.  If it weren't for our advertising efforts very few, if any, of our users would be aware of Bitcoin or making money in this way right now - that alone justifies not only a premium over the <10% profit margin of other pools, but it also justifies our expectation of being treated with decency.

[1714208017]

1714208017

[1714208017]

1714208017

[1714208017]

1714208017

[Compute4Cash]

@Grinder - the Compute4Cash client has a few auto-adjusting performance parameters that take a few minutes to settle down.  Once settled at full throttle it should mostly stay around 98%-99% if you are not doing anything else with your computer.

[m0mchil]

I have nothing against Compute4Cash.

Please just make a separate thread to promote it.

Thanks

[alikim]

I got this error, not sure if it's the connection problem or a program bug:

18/02/2011 23:54:48, Unexpected error:
Traceback (most recent call last):
  File "BitcoinMiner.pyc", line 210, in mine
  File "BitcoinMiner.pyc", line 187, in getwork
  File "httplib.pyc", line 974, in getresponse
  File "httplib.pyc", line 391, in begin
  File "httplib.pyc", line 355, in _read_status
BadStatusLine

[xenon481]

I got this error, not sure if it's the connection problem or a program bug:

18/02/2011 23:54:48, Unexpected error:
Traceback (most recent call last):
  File "BitcoinMiner.pyc", line 210, in mine
  File "BitcoinMiner.pyc", line 187, in getwork
  File "httplib.pyc", line 974, in getresponse
  File "httplib.pyc", line 391, in begin
  File "httplib.pyc", line 355, in _read_status
BadStatusLine

That is caused by the pool on the server side.

[Nonlin]

So as a 5870 user starting in a pool as of FEb 17. Would it be wise for me to go solo or just stay in the pool?

What would you Experience users suggest I do.

[ribuck]

So as a 5870 user starting in a pool as of FEb 17. Would it be wise for me to go solo or just stay in the pool?

You could join the pool at first, just so you know that your rig is working. Otherwise, it may go for days without generating, and you will find yourself wondering if it is working properly.

When you are happy that your rig is working, you can switch to solo mining to get 2% more coins, on average. With a 5870 you will average just over one block a week, but some weeks you will get 2 or 3 blocks, and sometimes you will go 2 or 3 weeks before you get a block. It's very variable.

[wolfangel91]

hi all,
I was on compute4cash but now I want to try bitcoin GPU mining. I read all this help http://www.newslobster.com/random/how-to-get-started-using-your-gpu-to-mine-for-bitcoins-on-windows#comment-14408
But I still have an error, I launch the bat for poclbm.exe and I see a serie of that in the console:
"incomplete read:incomplete read(296bytes read, 15 more expected) unexpected error traceback (most recent call at last):
File "bitcoinminer.pyc", line 210 in mine
File "bitcoinminer.pyc", line 188 in getwork
File "httplib.pyc", line 525 in read
File "httplib.pyc", line 617 in _safe_read"

Any idea ? what's that shit?  How to resolve this problem? thanks for help!

(and sorry for my bad english.. )

[m0mchil]

@wolfangel91

Disregard this error, it doesn't affect functionality at all. I just missed to handle and show server-side problem properly, will do so in next release this weekend.

[wolfangel91]

thanks for your response! you said "Disregard this error, it doesn't affect functionality at all", ok but when I look to GPU load (with AMD system monitor) it's show 2% to 12% load only.. so I think I doesn't work no?...But maybe it's my fault I may be incorrectly set something 

[Dude65535]

wolfangel91 make sure you have the right device selected and are running the miner on your gpu and not your cpu.

[Nonlin]

So now that I know my GPU is mining. To switch to solo do I just remove the --host line and address from my batch file and I will start mining for myself again?

[FooDSt4mP]

So now that I know my GPU is mining. To switch to solo do I just remove the --host line and address from my batch file and I will start mining for myself again?

You'll also have to make sure the username and password are correct, but you'll get an error right away if it can't connect.

[wolfangel91]

wolfangel91 make sure you have the right device selected and are running the miner on your gpu and not your cpu.

I'm sure, my CPU is from 2 to 4% and my GPU from 1 to 12%...so I think something doesn't work for me I've ATI SDK installed, drivers are up to date.. I see 8 connections on bitcoin and 108994bloks.. I've tried device 0 or device 1 and it's the same problem...

[urizane]

I found bitcoin.org from the Security Now podcast and I have some stats that may interest you guys.

My system:
Core i7-950 @ 3.72 GHz
6 GB of RAM
2x nVidia 1GB GTX 460s @ 810 MHz core 4000 MHz memory

Bitcoin client on CPU: 5.5 Mhash/s
System Load: 210 W

poclbm on 1 GPU: 56 Mhash/s
System Load: 210 W

poclbm on 2 GPUs: 112 Mhash/s
System Load: 320 W

poclbm on 2 GPUs + Bitcoin client on CPU: 117.5 Mhash/s
System Load: 420 W

Running the CPU hashing is not worth it for the energy cost of running the Bitcoin client's generator.  It also appears that, as it was said before, nVidia GPUs don't seem to run this OpenCL code very well.  I still get 10 times the performance out of the GPUs than I do out of the CPU, so there's that at least.  Maybe it has to do with the fact that I'm still running 260.89 for the driver.  I'm lazy, though, so I may not update it any time soon.  There may also be some losses due to the fact that Fermi parts that aren't the full size 3 billion transistor parts are superscalar, meaning that 2/3 of the shaders are guaranteed to work all the time and only ILP optimizations can make the other 1/3 do some good.  I'd be interested to see if any changes in the way the operations are structured could make it run any faster.  As far as I can tell, there's no way to verify that every last shader is doing work all the time with software available to the average consumer.  Afterburner is telling me that my GPUs are both 98%-99% busy.  From what I can see in the power readings above, the videocards are consuming roughly 110 W of power and 1GB GTX 460s are rated for 160 W.  I just thought that I would put that out there.

As a side note, with SLI enabled, I would still have to run two instances of poclbm to get both GPUs busy.  If I intend to do any sort of gaming while this is going on, it's better to turn off SLI and dedicate the secondary GPU to the task of generating and not have the occasional slowdown when the driver decides it wants to try and render a frame on the already taxed GPU.  Apparently with the driver set up the way it is, device 0 is my secondary card so I don't have to make any special considerations if I want to run just one GPU into the ground and game with the other.

All in all, great software.  I haven't gotten a block yet, but I've only been running it for 3 hours.  I also got set up by reading: http://www.newslobster.com/random/how-to-get-started-using-your-gpu-to-mine-for-bitcoins-on-windows

The only other time I've heard my system this loud is in Crysis and a mezzed up replay from Starcraft II.  Also, the wall behind my computer is significantly warmer than room temperature.  Maybe the GPUs are doing as much work as they can after all.  I just would have expected a higher system power load.

[xenon481]

I found bitcoin.org from the Security Now podcast and I have some stats that may interest you guys.

My system:
Core i7-950 @ 3.72 GHz
6 GB of RAM
2x nVidia 1GB GTX 460s @ 810 MHz core 4000 MHz memory

Bitcoin client on CPU: 5.5 Mhash/s
System Load: 210 W

I'd be interested in seeing the difference in CPU performance if you were to run this CPU Miner: http://bitcointalk.org/index.php?topic=3486.0

[robzy]

Hey guys, I'm afraid I'm having trouble with this. I did a quick test to make sure that bitcoind really was listening, with telnet, and it was.

I've double and triple checked the username and password.

Code:

[root@ip-*-*-*-* ~]# python2.7 poclbm.py  -d 0 -o 127.0.0.1 -p 8332  --user ***** --pass *****
Problems communicating with bitcoin RPC
bye
[root@ip-*-*-*-* ~]# telnet 127.0.0.1 8332
Trying 127.0.0.1...
Connected to localhost.localdomain (127.0.0.1).
Escape character is '^]'.
fhfsdhfd
sdghsdgsda
sdghsdhsdhas

HTTP/1.0 401 Authorization Required
Date: Sat, 19 Feb 2011 05:56:47 GMT
Server: bitcoin-json-rpc
WWW-Authenticate: Basic realm="jsonrpc"
Content-Type: text/html
Content-Length: 296

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
"http://www.w3.org/TR/1999/REC-html401-19991224/loose.dtd">
<HTML>
<HEAD>
<TITLE>Error</TITLE>
<META HTTP-EQUIV='Content-Type' CONTENT='text/html; charset=ISO-8859-1'>
</HEAD>
<BODY><H1>401 Unauthorized.</H1></BODY>
</HTML>
Connection closed by foreign host.

[root@ip-*-*-*-* ~]# 

Any suggestions on debugging the problem?

[edit]: Argh! It's "--pass" not "-p" Solved!

Thanks,
Rob.

[urizane]

I'd be interested in seeing the difference in CPU performance if you were to run this CPU Miner: http://bitcointalk.org/index.php?topic=3486.0

Holy bejeebus.  I guess the built-in one isn't so optimized.

Core i7-950 @ 3.72 GHz: 17.7 Mhash/s
System load: 185 W

I'm wondering why the power went down, though.

Also, adding -v to the OpenCL miner gave me 57.5 MHash/s as opposed to 56 MHash/s on a single 460.  Power consumption unchanged.

EDIT: CPU power difference solved.  Apparently the SSE2 miner makes my CPU "decide" to run at 3.57 GHz instead of 3.72 GHz.  No worries.  17.7 Mhash/s is much better than 5.5 Mhash/s anyway.  Maybe I'll lock the multiplier some day and stop listening to my CPU whining about the work...maybe.

[urizane]

I found the CUDA portion of puddinpop's miners and have a comparison of the two for nVidia folks.  AMD owners can safely ignore this post.

poclbm on both GPUs: 2x 57.5 Mhash/s (115 Mhash/s)
System load: 320 W

rpcminer-cuda on both GPUs: 2x 59.2 Mhash/s (118.4 Mhash/s)
System load: 305 W

I guess CUDA (being an nVidia thing) really does still work better on nVidia cards than OpenCL, but at least OpenCL is sooo damn close.  It's nice to see that on nVidia's own hardware, OpenCL is competitive.  It's just annoying that a pair of cards that do Crysis so damn well is outclassed in this application by a single 5770.  Oh well, back to games, I guess.

[sc8nt4u]

Hey m0mchill, was -v command changed in this update from 0205? My .bat files that worked for 0205 does not work for 0215. It will only work if I take out the -v from the .bat file, but i lose about 30K/hash per card.