Ubuntu Natty Narwhal 11.04 Mining Guide / HOWTO

[mintymark]

Alas, MOA, this resultaed in an unbootable configuration. aticonfig did find the single remaining device instance though, and the ordering of screens etc was slightly different.
Will work on reducing the screens etc to try to get bootable system as suggested.

[1714328872]

1714328872

[1714328872]

1714328872

[1714328872]

1714328872

[1714328872]

1714328872

[mintymark]

Ok, it seems that it really does not like more than one BusId statement, this causes failure: X will not boot. However if the BusId is the same it does boot, but still does  gives rise to just 1 Caymen device for use by poclbm.

I think 6990 cards provide 3 devices at 3 different PCI address at boot, a VGA adapter (for use by the bios etc) , an audio adapter and an ATI GPU device. Subsequently I think they reuse the VGA graphics adapter to add the 2nd GPU, this may be quite wrong of course.

I think that X, thinking there is a VGA adapter at the 2nd location doesnt know what to do.

This translates as, its broken.

Somebody with 6990 cards working like to comment?

[mintymark]

Is also hashing at only 100 Mh/sec

[aeroSpike]

I have managed to set this up. I have 2 cards, a 5850 and a 5870. But I only have 1 monitor connected.

Code:

aticonfig --list-adapters

shows my 2 cards, but only 1 is visible with the other commands such as:

Code:

aticonfig --odgc --adapter=all

I suspect this is because one of my cards is disabled because it has no monitor connected. I tried editing /etc/X11/xorg.conf and add a second "Device" section, but to no avail.
Anybody with a suggestion here? Do I really have to buy a second monitor?

EDIT: hmm. I reran:

Code:

sudo aticonfig --initial -f --adapter=all

I rebooted and now both my boards are active. Apparently manually editing /etc/X11/xorg.conf and copying the "Device" Section (setting the PCI slot to the correct id) caused aticonfig to create the missing sections. After restarting X I am now mining with 2 boards yet using one single monitor.

Hope this helps for others who have the same problem.

[mjhorn]

Are you sure you have the SDK installed properly?  Might try going back and doing it again.

Which file did you install?

I appreciate the help so far. I have tried to install process a few times for the SDK, but always with the same results. I'm using the 64bit SDK from http://download2-developer.amd.com/amd/APPSDK/AMD-APP-SDK-v2.4-lnx64.tgz. For the "install" process, I've unpacked to /opt, set the associated environment variables:

export AMDAPPSDKROOT=/opt/AMD-APP-SDK-v2.4-lnx64/
export AMDAPPSDKSAMPLESROOT=/opt/AMD-APP-SDK-v2.4-lnx64/
export LD_LIBRARY_PATH=/opt/AMD-APP-SDK-v2.4-lnx64/lib/x86_64/:/lib:/usr/lib:/usr/local/lib
export ATISTREAMSDKROOT=/opt/AMD-APP-SDK-v2.4-lnx64/

export DISPLAY=:0


It seems to be working to at least some degree as running clinfo does detect both of my CPU cores, just no GPUs. Also, I've noticed that if I try running clinfo as root with sudo, it does error out with:

/opt/AMD-APP-SDK-v2.4-lnx64/bin/x86_64/clinfo: error while loading shared libraries: libOpenCL.so.1: cannot open shared object file: No such file or directory

However, that libOpenCL.so.1 does exist and is found properly when run as a standard user.

[mintymark]

I am at a similar stage to mjhorn.

I have started to focus less on the xorgfile, leaving out all bus id statements, and more on PyOpen and the sdk.

I see 1 Cayman device only.

Have moved to sdk 2.3. I still see only 1 Cayman device. For a period of 2 days I had 2 of my 3 6990's running (thats 4 devices) under 2.3. I do not seem to be able to achieve even that now.

CLinfo states there is one GPU and one CPU, effectively agreeing with poclbm.

I have become addept at changing sdk version, but I wonder if there is something missing when doing a reinstall of the sdk. This is what I do:

#!/bin/bash
rm siteconf.py
sdk=/opt/ati-stream-sdk-v2.3-lnx32

export LD_LIBRARY_PATH=$sdk/lib/x86
./configure.py --cl-inc-dir=$sdk/include/ --cl-lib-dir=$sdk/lib/x86
make clean
make -j3
sudo make install

And then

check that the name of the library in /etc/OpenCL/vendors/atiocl32.icd agrees

and then edit .bashrc for the sdk name which currently is

export LD_LIBRARY_PATH=/opt/ati-stream-sdk-v2.3-lnx32/lib/x86/:$LD_LIBRARY_PATH

Wondering if DISPLAY=:0 is correct.

[mintymark]

Changing sdk hasnt helped me much, sdk 2.3, and 2.4 both see 1 Cayman gpu. 2.1 none at all.

This is the script I use to swop sdk's:

#!/bin/bash
sdk=/opt/ati-stream-sdk-v2.4-lnx32

export LD_LIBRARY_PATH=$sdk/lib/x86
cd ~/pyopencl-0.92
rm siteconf.py
./configure.py --cl-inc-dir=$sdk/include/ --cl-lib-dir=$sdk/lib/x86
make clean
make -j3
sudo make install
ls -1S $sdk/lib/x86 | head -1 > /tmp/atiocl32.icd
sudo cp /tmp/atiocl32.icd /etc/OpenCL/vendors/atiocl32.icd
rm /tmp/atiocl32.icd
grep LD_LIBRARY_PATH ~/.bashrc | grep -v ati-stream > /tmp/bashrc && cp /tmp/bashrc ~/.bashrc && rm /tmp/bashrc
echo "export LD_LIBRARY_PATH=$sdk/lib/x86:$LD_LIBRARY_PATH" >> ~/.bashrc

[Inaba]

I'm sorry you guys are having so much trouble.  I'm working on acquiring a 6990 or two and I'll be able to work it out.  It's hard to get a hold of those beasts at a reasonable price.  Hopefully I can get one soon...  I did have one briefly and had some problems, but since I was selling it anyway, I didn't bother to look deeper into the issue.

[mintymark]

What town (and country!) are you in Inaba?

[Inaba]

I live in Kansas City, KS

[mjhorn]

Minty, just curious, what brand are your 3 6990s? Wondering if that might possibly have anything to do with whether or not the cards are detected properly. For what it's worth I've got the Diamond version.

My only other thought is that at least in my case, I started with what I originally had built as a server/network storage box, so it was only accessible via ssh with no X components installed at all. I've done a pretty minimal X install, but I'm wondering if there are still components I'm missing since I had a non-standard starting point with the Ubuntu install. I think my next step is going to be a fresh install, I'm just a bit hesitant to do so as I do have enough other things running on here currently that it would be quite a hassle to properly back up and reinstall them as well.

[darvil]

I want to say I really appreciate your thread.  Helped me alot actually.  Got quite a few cards running now.

I'll send you some coins.  Starting off with 1 first

[ryepdx]

I've been having similar problems getting my 3 cards working. I have two 5970s and a 6990. My issue is identical to the one minty had initially. I see the Ubuntu logo, and then nothing. The monitor (plugged into the 6990 in PCIE_1) gets a signal, though, and if I plug it into the second 5970's top DVI socket I get a "Signal Out Of Range" error on the monitor. :-/

[mintymark]

If you could only see the amount of hair I have left, you would be amazed: I have been tearing it out. But I think I have this figured out now, or at least "a method". The brilliance is not in fact mine.

I had brought 3 cards, and spent days getting nowhere. I was actually thinking of trying to return them because I was SOOO fed up, the did not after al;l seem to do what it said on the tin, which is work with Linux. I mean not in any simple way, and it all the worse for the fact that some people have it working, with, say the catalyst driver 11.3 which seems to have an installer that is broken on Ubuntu 11.04. And to even get X to accept the xorg.conf, I had to rtake out most of the busid statements, which are supposed to be mandatory. Aaaagh!.

The solution is Linuxcoin, in this thread. This is a bootable iso image: its not Ubuntu, but it is linux, and everything is installed, and it runs all 6 cores of my 3 6990 cards.

Of course it would be nice if the power supply did not overheat and blimp of with no warning, but that is another issue!!

http://forum.bitcoin.org/index.php?topic=7374.0

[ryepdx]

The solution is Linuxcoin, in this thread. This is a bootable iso image: its not Ubuntu, but it is linux, and everything is installed, and it runs all 6 cores of my 3 6990 cards.

Meh. This doesn't help me. I tried Linuxcoin and couldn't get it to boot. :-/

Thanks though.

[mjhorn]

Well, finally able to get things running. After yet another wiping of the drivers and SDK, another reinstall is working, at least when run as root. Still unable to detect the cards as a normal user account however, but this seems like it should be much easier to fix. Any ideas where the permissions might be goofy that clinfo is able to run, but only detects the card as root?

[MiningBuddy]

Used this guide to setup my 4x gpu machine earlier today and I haven't used Ubuntu since version 6.
Everything went perfect and I was mining within a couple of hours.

Thank you very much for taking the time to write and support this guide.

[IlbiStarz]

Poclbm isn't seeing my 6870's. How do I install 2.4?

[marcus_of_augustus]

When running ./bitcoin -server, it tells me I need to create a .conf file with the rpc stuff, which I did.
I put it in home/nick/bitcoin/ is that correct or do I put it somewhere else?

AFAIK about your set-up, it should go in /home/nick/.bitcoin

with the block data and the wallet.dat

[IlbiStarz]

Okay I finally got everything to work.

When I do ./poclmb.py -o btcguild.com -u (username_user) --pass=(password) -p 8332 -d 1, it gives me problem connecting with bitcoin error. It'll stay there for about a minute, then it'll start mining. But then 5 seconds after that, it'll say "verification failed, check hardware!"

Help!! So close!!