50 BTC if you write a complete guide on GPU mining on Ubuntu using ATI

[mintymark]

Yes, crossfire off, and the connector between the boards not connected.

[1714080082]

1714080082

[Raulo]

So far both the ATI cards work and give the expected hashrate, but if I run 2 threads then the combined hashrate is lower than with one card.

Do you run poclbm with different device numbers, e.g. "-d 1" for the first GPU and "-d 2" for the other (adjust the numbers according to the list printed by poclbm with no options)? It looks like you don't and if you run it with the same device, they will compete for the graphics card.

By the way (but do it after you solve the above issue), SDK 2.1 is faster.

[mintymark]

Yes thats what I did.

[Raulo]

Yes thats what I did.

Well, let's check what you did because it should work.

Could you write the output of ./poclbm.py without any options
and your command line options for poclbm.py (obviously without user and pass)?

[mintymark]

Thank you! Obviously I HAVE made some error because it SHOULD work. Here goes:

export DISPLAY=:0
mark@bigdaddy:~$ ./poclbm.py
No device specified or device not found, use -d to specify one of the following

[ 0]   Cypress
[1]   Cypress
[2]   Intel(R) Core(TM) i7 CPU         950  @ 3.07GHz
./poclbm.py -f 100 --verbose -w 128 -r 60 -v -d1 --user ...  --pass ...

-d0  works exactly the same way so far as I can see.
I use -f 100 because it makes only a small difference to the hash rate but increases display responsiveness a lot.

[Raulo]

./poclbm.py -f 100 --verbose -w 128 -r 60 -v -d1 --user ...  --pass ...

If you run the above in one window and

Code:

./poclbm.py -f 100 --verbose -w 128 -r 60 -v -d0 --user ...  --pass ...

in another (notice different device numbers) and you still have problems, then it's weird. Do it and show the output of

Code:

aticonfig --odgc --adapter=all

Both cards should have close to 100% usage.

You may also try SDK 2.1 because SDK 2.3 may be broken and SDK 2.1 is faster so it's best to use it anyway.
Notice that the device numbers change with SDK 2.1.

I use -f 100 because it makes only a small difference to the hash rate but increases display responsiveness a lot.

If by "small difference" you mean about 10% then I agree (in my opinion 10% is quite a lot if you don't care about responsiveness). 

[mintymark]

1st window:

mark@bigdaddy:~/poclbm20110311$ ./poclbm.py -f 100 --verbose -w 128 -r 10 -v -d0 --user ... --pass ...
25/03/2011 23:04:05, 187777 khash/s
25/03/2011 23:04:11, checking 3512679190 <= 62265
25/03/2011 23:04:15, 186506 khash/s
25/03/2011 23:04:20, checking 3742602603 <= 62265
25/03/2011 23:04:25, 189006 khash/s
25/03/2011 23:04:35, 192135 khash/s
25/03/2011 23:04:45, 190200 khash/s
25/03/2011 23:04:55, 189390 khash/s
25/03/2011 23:05:05, 193691 khash/s
25/03/2011 23:05:15, 194551 khash/s
25/03/2011 23:05:22, checking 3736923992 <= 62265
25/03/2011 23:05:25, 192253 khash/s
25/03/2011 23:05:31, checking 1440968788 <= 62265
25/03/2011 23:05:35, 190937 khash/s
25/03/2011 23:05:45, 193015 khash/s
^C
Compilation terminated.

2nd window:
mark@bigdaddy:~/poclbm20110311$ ./poclbm.py -f 100 --verbose -w 128 -r 10 -v -d1 --user ... --pass ...
25/03/2011 23:04:06, 85915 khash/s
25/03/2011 23:04:16, 78925 khash/s
25/03/2011 23:04:26, 80097 khash/s
25/03/2011 23:04:36, 80161 khash/s
25/03/2011 23:04:46, 79075 khash/s
25/03/2011 23:04:56, 75824 khash/s
25/03/2011 23:05:06, 73409 khash/s
25/03/2011 23:05:16, 75611 khash/s
25/03/2011 23:05:26, 74499 khash/s
25/03/2011 23:05:36, 74750 khash/s
25/03/2011 23:05:46, 74164 khash/s
25/03/2011 23:05:54, checking 3327463497 <= 62265
25/03/2011 23:05:56, 241692 khash/s
25/03/2011 23:06:02, checking 589964444 <= 62265
25/03/2011 23:06:06, 255866 khash/s
25/03/2011 23:06:06, checking 2921145210 <= 62265
25/03/2011 23:06:11, checking 2796295684 <= 62265
25/03/2011 23:06:16, 256471 khash/s
^C
Compilation terminated.

3rd window:

mark@bigdaddy:~$ date ; aticonfig --odgc --adapter=all
Fri Mar 25 23:05:24 GMT 2011
Adapter 0 - ATI Radeon HD 5800 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    765           1125
             Current Peak :    765           1125
  Configurable Peak Range : [550-900]     [900-1300]
                 GPU load :    72%

Adapter 1 - ATI Radeon HD 5800 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    765           1125
             Current Peak :    765           1125
  Configurable Peak Range : [550-900]     [900-1300]
                 GPU load :    28%
mark@bigdaddy:~$ date ; aticonfig --odgc --adapter=all

Fri Mar 25 23:06:10 GMT 2011
Adapter 0 - ATI Radeon HD 5800 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    157           300
             Current Peak :    765           1125
  Configurable Peak Range : [550-900]     [900-1300]
                 GPU load :    0%

Adapter 1 - ATI Radeon HD 5800 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    765           1125
             Current Peak :    765           1125
  Configurable Peak Range : [550-900]     [900-1300]
                 GPU load :    96%
mark@bigdaddy:~$ date ; aticonfig --odgc --adapter=all

Fri Mar 25 23:07:19 GMT 2011
Adapter 0 - ATI Radeon HD 5800 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    765           1125
             Current Peak :    765           1125
  Configurable Peak Range : [550-900]     [900-1300]
                 GPU load :    0%

Adapter 1 - ATI Radeon HD 5800 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    157           300
             Current Peak :    765           1125
  Configurable Peak Range : [550-900]     [900-1300]
                 GPU load :    0%


Explanation:

What I did here is broadly what you suggested. windows 1 and 2 were started as near as I could at the same time.
Although both rates are low, the 2nd window is lower. I stop the 1st window and the 2nd picks up in speed.

In the 3rd window, I run the command a) when both threads are running. b) after the first has been stopped c) after both are stopped.


Any thoughts?

[Raulo]

Any thoughts?

You are running it correctly but the threads apparently compete for the same GPU instead of running on their own. 

My ideas:

1. Buggy SDK
2. Buggy Catalyst driver (10.11 or 10.12 are reported to work best). What's your version?
3. Broken xorg.conf

Could you show your "Device" sections in /etc/X11/xorg.conf?

[mintymark]

Catalyst version 10.12 according to amdcccle.

I have always hated Xorg.conf!  Here are my device sections:

Section "Device"
    Identifier  "aticonfig-Device[0]-0"
    Driver      "fglrx"
    BusID       "PCI:3:0:0"
EndSection

Section "Device"
    Identifier  "aticonfig-Device[1]-0"
    Driver      "fglrx"
    BusID       "PCI:4:0:0"
EndSection

And while on the subject, here is
lspci | grep ATI
03:00.0 VGA compatible controller: ATI Technologies Inc Cypress [Radeon HD 5800 Series]
03:00.1 Audio device: ATI Technologies Inc Cypress HDMI Audio [Radeon HD 5800 Series]
04:00.0 VGA compatible controller: ATI Technologies Inc Cypress [Radeon HD 5800 Series]
04:00.1 Audio device: ATI Technologies Inc Cypress HDMI Audio [Radeon HD 5800 Series]

Regards, and thanks for your help !

[Raulo]

Catalyst version 10.12 according to amdcccle.

I have always hated Xorg.conf!  Here are my device sections:

Looks OK. If you have also correct  "Screen" sections (pointing to different devices) then the only thing is SDK 2.3. I read reports that it had some problems with multi-GPU settings:
http://forums.amd.com/forum/messageview.cfm?catid=390&threadid=143851

Install SDK 2.1 and cross fingers. After reading the above thread, it's probably the case.

edit: Apparently

Code:

export GPU_USE_SYNC_OBJECTS=1

should also help (you can make a quick test) but there is no reason to use SDK 2.3.

[mintymark]

Screens section is as follows:
Section "Screen"
    Identifier "aticonfig-Screen[0]-0"
    Device     "aticonfig-Device[0]-0"
    Monitor    "aticonfig-Monitor[0]-0"
    DefaultDepth     24
    SubSection "Display"
        Viewport   0 0
        Depth     24
    EndSubSection
EndSection

Section "Screen"
    Identifier "aticonfig-Screen[1]-0"
    Device     "aticonfig-Device[1]-0"
    Monitor    "aticonfig-Monitor[1]-0"
    DefaultDepth     24
    SubSection "Display"
        Viewport   0 0
        Depth     24
    EndSubSection
EndSection


I hear what you say, I will try 2.1 I guess.

Thanks very much for your help.

[mintymark]

I was in truth rather discouraged by all this. Last time I had tried to change versions it was very difficult to get it working again at all.

But the export GPU_USE_SYNC_OBJECTS=1 did it, with 2.3, no need to re-install 2.1

So thanks very much for that clue!!

[pawel]

It's any way to monitoring GPU and CPU temperatures with Bitcoing progress of work by telnet ? Or i obliged to use remote desktop :/

[Mahkul]

It's any way to monitoring GPU and CPU temperatures with Bitcoing progress of work by telnet ? Or i obliged to use remote desktop :/

Try:

While logged in via ssh try this:

Code:

aticonfig --odgt --adapter=all

If you get errors do:

Code:

export DISPLAY=:0

And then try aticonfig again.

EDIT: By the way, there is a Polish bitcoin forum at http://bitcoin.pl/forum (I assume you're Polish judging by your nick).

EDIT 2: As for the CPU try typing:

Code:

sensors

If this doesn't work follow this guide.

[blap]

I got:

Code:

Traceback (most recent call last):
  File "poclbm.py", line 3, in <module>
    import pyopencl as cl
  File "/usr/lib/pymodules/python2.6/pyopencl/__init__.py", line 3, in <module>
    import pyopencl._cl as _cl
ImportError: libOpenCL.so.1: cannot open shared object file: No such file or directory

And I do not kown why, but

Code:

$ ./CLInfo |grep CL_DEVICE_TYPE_GPU

got "No such file or directory".
=(

[Raulo]

I got:

Code:

Traceback (most recent call last):
  File "poclbm.py", line 3, in <module>
    import pyopencl as cl
  File "/usr/lib/pymodules/python2.6/pyopencl/__init__.py", line 3, in <module>
    import pyopencl._cl as _cl
ImportError: libOpenCL.so.1: cannot open shared object file: No such file or directory

Code:

$ ./CLInfo |grep CL_DEVICE_TYPE_GPU

got "No such file or directory".

Both errors are because SDK is not installed or installed in a different directory than in the guide.

[blap]

I got:

Code:

Traceback (most recent call last):
  File "poclbm.py", line 3, in <module>
    import pyopencl as cl
  File "/usr/lib/pymodules/python2.6/pyopencl/__init__.py", line 3, in <module>
    import pyopencl._cl as _cl
ImportError: libOpenCL.so.1: cannot open shared object file: No such file or directory

Code:

$ ./CLInfo |grep CL_DEVICE_TYPE_GPU

got "No such file or directory".

Both errors are because SDK is not installed or installed in a different directory than in the guide.

ok, I used sdk 2.4, but

Code:

export LD_LIBRARY_PATH=/opt/ati-stream-sdk-v2.4-lnx64/lib/x86_64/:$LD_LIBRARY_PATH

will not link?
because I did not see in "/opt/ati-stream-sdk-v2.4-lnx64/lib/x86_64/" the folder "CLInfo" or something like this...

UPDATE:
I really need v 2.1? How I delete v 2.4?

[Raulo]

The guide was written for SDK 2.1 which is still the best choice for 5xxx cards. I don't have 2.4 but apparently AMD renamed some dirs and moved some programs.

The library path is AMD-APP-SDK-v2.4-lnx64/lib/x86_64 and clinfo (with small letters) is in AMD-APP-SDK-v2.4-lnx64/bin/x86_64

But if you want to continue with SDK 2.4, you are on your own because I didn't test it on this version.

[blap]

The guide was written for SDK 2.1 which is still the best choice for 5xxx cards. I don't have 2.4 but apparently AMD renamed some dirs and moved some programs.

The library path is AMD-APP-SDK-v2.4-lnx64/lib/x86_64 and clinfo (with small letters) is in AMD-APP-SDK-v2.4-lnx64/bin/x86_64

But if you want to continue with SDK 2.4, you are on your own because I didn't test it on this version.

ok, I found clinfo in "/opt/AMD-APP-SDK-v2.4-lnx64/samples/opencl/bin/x86_64", but

Code:

$ cd /opt/AMD-APP-SDK-v2.4-lnx64/samples/opencl/bin/x86_64
$ ./clinfo |grep CL_DEVICE_TYPE_GPU

did not work too...
I will try v2.1...

[Mahkul]

The guide was written for SDK 2.1 which is still the best choice for 5xxx cards. I don't have 2.4 but apparently AMD renamed some dirs and moved some programs.

The library path is AMD-APP-SDK-v2.4-lnx64/lib/x86_64 and clinfo (with small letters) is in AMD-APP-SDK-v2.4-lnx64/bin/x86_64

But if you want to continue with SDK 2.4, you are on your own because I didn't test it on this version.

ok, I found clinfo in "/opt/AMD-APP-SDK-v2.4-lnx64/samples/opencl/bin/x86_64", but

Code:

$ cd /opt/AMD-APP-SDK-v2.4-lnx64/samples/opencl/bin/x86_64
$ ./clinfo |grep CL_DEVICE_TYPE_GPU

did not work too...
I will try v2.1...

It's ./CLInfo not ./clinfo

Unless you were just lazy in your post.