Ubuntu Natty Narwhal 11.04 Mining Guide / HOWTO

[minerX]

Not true.  There are a couple different tools that allow Linux overclocking, which have already been mentioned in this thread.

Personally, I flash the BIOS with RBE and don't even worry about overclocking from the command line.  If you can't overclock in Linux, the fault lies with the operator, not with the (lack of) ability.  Is it as slick and easy as Windows?  Nope, but then again, Windows can't do half the things a Linux machine can in terms of mining, so it's a trade off.  I'd like to see you run more than 8 GPUs in Windows.  Can you even run more than 4?

My ASUS 5850 DirectCU 725 are virtually impossible to overvolt.   GPU-Z cannot pull the bios from the cards.   So I instead download a bios version that looks similar from guru3d.   The bios works fine.  After that I attempt to edit via RBE.  No dice.  Once I edit the voltage the GPUs are no longer recognized by aticonfig and/or windows.   They are "greyed" out.   Perhaps its user error in the RBE editor, but I sure as shit can't figure it out.

[1713967360]

1713967360

[1713967360]

1713967360

[1713967360]

1713967360

[1713967360]

1713967360

[Inaba]

Im going to disagree here just on the side of mining though.  I am a Redhat certified engineer (Earned RHCE earlier this year), and I work with RHEL6 and RHEL5 machines for a career. So I totally understand the subtleties between Windows and Linux.
But, there is absolutely nothing available that will allow you to overvolt a a non-reference GPU in linux. There are a multitude of tools for Windows that allows voltage adjustements for non-reference cards.  Trixx, MSI Afterburner are just a few.  AMDovrdrvctrl will not allow you to adjust voltages on non-reference cards.  You also have to modify the BIOS ont the GPU if you want to do anything worth while with AMDovrdrvctrl.  You dont have to modify the BIOS to do the same with the Windows tools mentioned above.

The complaint was there was no way to overclock in Linux and that complaint is false. There was no mention of overvolting.  Regardless, though, overclocking and overvolting via software is a ludicrous proposition anyway for anything but a temporary measure. I've never understood the trend, honestly.  You get shitty results and a bucket load of instability in the long term when you do it via software, this applies to both Windows and Linux.  The only thing software overclocking is good for is to find your stable numbers.  Once you find those, you should flash your card with those and leave it at that.  Software tweaking is a temporary measure and for the purposes of mining, it's completely pointless.  If you're going to mine you need your cards to be set to what you're mining with.  If you're some hobbiest miner that is only doing it now and again while using the computer for other tasks, then you're probably not going to be using Linux anyway.  If you're following this guide then you most likely are serious about mining and have a dedicated box and thus using software to overclock for mining is just wrong on so many levels.

For example, I can overclock all 4 of my GPU's in Windows and they will stay STABLE at higher clock speeds than in ubuntu.  In windows I Can run my 3 HD5850's at 950 core 325 RAM at 1.188 volts and they will hash at 360-370 Mhash all day long, below 65-70c.  I can also overclock an additional HD5870 in this same rig to 1000 core 350 ram at 1.2volts and it will stay stable.

I have 28 miners running, all of them overclocked in under Ubuntu and the only ones that have stability problems are the 5970's and that's due to some other issues unrelated to overclocking (craptacular VRM cooling mostly).  All the 5870's are rock solid at 950/300 with 5 - 8 GPUs per motherboard.  There's a problem somewhere else if you are unable to achieve stability only in Linux.  Although, I admittedly don't have nor have I used any 5850's, I can't imagine them being much different.

Now, in Ubuntu Natty, 64 bit running SDK 2.4 and Catalyst 11.5 I CANNOT get the stupid GPU's to run for over an hour at 800 core and 300 ram.  With these clock speeds  I only see about 300 Mhash/sec. These GPU's have the same BIOS settings as they do when I can overclock them in Windows.   I have read every post in this thread, and posted questions in several others with no improvement in results.

There is something very wrong with your setup if you're only getting 300 MH/s at 800 core.  You should be getting at least 350, which is what I get on stock core on the 5870's.  Seeing as you're running 2.4 that probably doesn't help and you should at least be using Phoenix with phatk kernel if you're hell bent on using 2.4 with 5xxx series cards.  In either case, moving to v2.1 if you only have 5xxx series cards will improve your hash rate dramatically.

You yourself even mentioned you had no idea how to free a froze GPU in linux.  I have not found a way to kill the process or even detach the module from the kernel without causing a kernel panic.  I was highly dissappointed to find out that I can run my 3 5850's at 950 and my 5870 at 1000 for days on end in Windows without the first freeze.  Linux has never had a great relationship with GPU's (especially ATI).

Absolutely, but I also had the same problem in Windows (which is why I moved to a Linux solution in the first place) - the GPU would freeze and there'd be no way to unstick it without a full power cycle in Windows.  The same problem exists in Linux, so there's no advantage there.  However, the major advantage with Linux that you don't have with Windows is the fact that I can access the machine through SSH and take care of everything - can't do that with Windows.  You have to have a convoluted, slow ass solution like Teamviewer, VNC, LogMeIn or similar just to manage your miners.  Monumental pain in the ass.

Please dont get me wrong, I am not trying to minimize what you are doing by helping out in this thread, but I have had no luck in getting the same results in Ubuntu as I have in WIndows, and I absolutely hate that.  This mining rig is actually the only Windows machine in my house (I am running Fedora and RHEL Workstation on my personal machines).  I have read every post, and tried everything in this thread as well as others. I still cannot get my GPU's stable above 800 core in ubuntu.  I am going to try Debian Squeeze 32 bit later tonight, and I may even try a fresh install of Ubuntu Maverick or Lucid since I despise Natty so much.

Then there is something wrong with your setup. I can't believe the 5850's are that much different than the 5870's and after flashing the BIOS of all the 5870's for 950 core / 300 mem, I've had zero problems with the 5870's.  Most of them don't even require a voltage tweak, though some of them I've had to bump the voltage a couple notches.

Not true.  There are a couple different tools that allow Linux overclocking, which have already been mentioned in this thread.

Personally, I flash the BIOS with RBE and don't even worry about overclocking from the command line.  If you can't overclock in Linux, the fault lies with the operator, not with the (lack of) ability.  Is it as slick and easy as Windows?  Nope, but then again, Windows can't do half the things a Linux machine can in terms of mining, so it's a trade off.  I'd like to see you run more than 8 GPUs in Windows.  Can you even run more than 4?

My ASUS 5850 DirectCU 725 are virtually impossible to overvolt.   GPU-Z cannot pull the bios from the cards.   So I instead download a bios version that looks similar from guru3d.   The bios works fine.  After that I attempt to edit via RBE.  No dice.  Once I edit the voltage the GPUs are no longer recognized by aticonfig and/or windows.   They are "greyed" out.   Perhaps its user error in the RBE editor, but I sure as shit can't figure it out.

If I can find a 5850 for cheap, I'll pick one up and see if it's that much different than the 5870's.  If you want to send me a copy of your stock BIOS I can take a look at it.

[CubedRoot]

Thanks for the replies Inaba.  I dont disagree that I have to have something wrong in my system.  I have swapped in a fresh HDD and installed a fresh copy of Debian Squeeze 32 bit.  I am about to go through the process of setting up Catalyst 11.5 and I think will try AMD SDK 2.1 this time. I have read on many different posts that SDK 2.4 when combined with Phoenix and PhatK are about the best for hashing on 5800 series GPUs, but its worth a shot to try 2.1 since you are getting good results with it.

If I could find the clocks these cards are stable at I would flash them in the BIOS. I just have not been able to get these cards stable in Linux yet. 

If you would like to take a look at the BIOS's for these cards I would much appreciate it.  These are the Stock Bios's that I used GPUz to pull from the cards before I done any tweaking on them.

Gigabyte HD5870 Stock BIOS:
http://dl.dropbox.com/u/417019/Gigabyte%20HD5870%20Stock%20bios.bin

Sapphire HD5850 Stock BIOS:
http://dl.dropbox.com/u/417019/Sapphire%205850.bin

Would you mind giving me a copy of your 5870 bios so that I can give it a shot here on my 5870?

This rig is destined to be a dedicated Miner if I can ever get it stable.

I only get about 300 to 310 Mhash/sec on 800 core 300 ram on my 5850's. You will probably see more than that on your 5870.

[Inaba]

I have a couple 2.4 systems as test systems running on 5870's.  The absolute best I've been able to achieve with 2.4 and phatk is just slightly less than "stock" 2.1 on poclbm.py.

I prefer poclbm.py anyway, since it seems to be a bit more stable than Phoenix when there's an RPC problem (it recovers better, though they both suck mightily at reliability when an RPC connection error occurs).  At this time, I don't see any reason at all to run 2.4 if you don't have 6xxx series cards, as it does nothing but hinder you or at best, it's at parity but requires some very specific combinations of software and config variables.  Whereas with 2.1 you can just slap it on there and go with what you know.

Here's a link to my 5870 BIOS that is rock solid at 950/300 on my XFX, Gigabyte and Sapphire 5870's:
http://dl.dropbox.com/u/1308992/soc_mine.rom

Of course, YMMV, but I've had no problems flashing it onto all 20+ of my 5870's.  I can't remember if I tweaked the voltages on that or not.  Bear in mind, it will overheat your card if it's thermally on edge already.  If you have good airflow/cooling though it hovers around 70C, at least on my cards. 

As far as GPU-Z goes, that may be the problem.  I don't know, as I've not used it. I use RBE in DOS to get the BIOS off my cards and to flash them onto the cards.  Being old school I have an innate fear of flashing anything in Windows. 

Your fan profile should be adjusted to spin up much faster.  Your voltages are llllow for overclocking.  You can be low by fractions of a tenth, but you're almost a whole tenth of a volt low for anything above 825 or so.

For the 5850 BIOS, your GPU registers 0x46 voltage doesn't match Mode 3's voltage.  I'm not sure if that's a problem or not, but I always make mine match.

For both BIOS, I've found that playing with memory clock from within a booted OS always causes a crash of some sort.  If I flash 300 on the cards, I've never had any memory related instability.

[supa]

Anyone suggesting that it's "impossible" to get the same numbers in Linux as they get in Windows....

.... is correct.  Linux numbers are much higher.

People have done an exceptional job of providing the answers - just use RBE.  AMDOverDriveCtrl doesn't give me problems on my cards.  Driver version is a definite possibility - anyone with any Linux experience should know that *generally* the drivers for Linux are much lower quality.  Whether they just be outright buggy or the installation is torture, they just simply aren't the same quality as Windows counterparts.

Here's a case study for you -

Several Diamond 5850s (non-reference, no voltage adjustments possible)
Flashed with FreeDOS to edited BIOS from RBE
Watchdog hardware module
100% unattended boot to init 3, xinit, AMDOverDriveCtrl profile load, miner startup
8 second boot time
Can't clock over 1005 MHz CPU without causing a kernel panic/hard reset
Manageable with screen/ssh
Catalyst 2.4
HAL independent - clonable to any machine
394 MHs on average with phoenix/phatk and aggression > 15.

*Exact* same card in Windows 7x32
Flashed same way (same card - did a "burn in" test in Linux before transferring it to Windows)
No watchdog.
Autologon + startup script for miner
37 second boot time
Can't clock over 925 MHz CPU without causing a driver fail, machine reset or poclbm/phoenix hanging
Aero turned off (no idea if this makes a difference....)
Manageable with.... VNC or garbage RDP
Catalyst 2.4
HAL dependent - lengthy install required
310 MHs on average with phoenix/phatk and aggression < 13 (otherwise it hangs)

If any Windows Pros want to tell me how to fix the Windows machine - fire away.

Also - what tool lets you set profiles and load them in a script like AMDOverDriveCtrl ?

PS - Fedora x32 and Fedora x64.  Miners aren't worth RHEL licenses.

[minerX]

If I can find a 5850 for cheap, I'll pick one up and see if it's that much different than the 5870's.  If you want to send me a copy of your stock BIOS I can take a look at it.

I didn't know you could RBE and pull bios through DOS.  I thought GPZ was the only way.  So my model right now is DIRECTCU 725.   However I flashed them all with the "765" version.   I have access to the "stock" through techpowerup.   However there is no way to be certain that the "stock" bios is compatible with my card (meaning they might have changed the card.)     So, here is the "stock 765" bios I'm using on all my cards.    

http://www.techpowerup.com/vgabios/75141/Asus.HD5850.1024.100329.html


This is what I think was the original stock one.  

http://www.techpowerup.com/vgabios/74071/Asus.HD5850.1024.100310.html


Here are my RBE edited ones(from the first link( that do not allow overclocking in windows or ATIconfig.  "greyed out, but still mineable")  I tried editing both the voltages and the highest step MHz.

** These were either edited from the 725 or 765 link noted above.  Forgot which, but they all end up the same. **

-- 1.3 volts only
http://dl.dropbox.com/u/8045114/585013voltonly.rom  

--1.3 volts, 1000 MHZ
http://dl.dropbox.com/u/8045114/5850_1_3volt1000

[chungenhung]

is there any steps that I need to re-run if I add in another GPU?
Right now there are 4 GPUs, I want to add another GPU to it.

[minerX]

is there any steps that I need to re-run if I add in another GPU?
Right now there are 4 GPUs, I want to add another GPU to it.

Run the initialize line again.


Also, I'm retarded and installed using 2.1 instead of 2.4.     Any easy way to swap over?  Or do I need to reinstall.

Thanks

EDIT:   Installed 2.4 and am using --platform 0.      Seems to be working OK.   

[minerX]

Ok,  I'm seriously getting irritated now.

I just reinstalled Ubuntu and am still having overclock issues.   When I overclock using ATIConfig,  my hashrates DECREASE.    My 6970 goes from 370 to 320.     

Could use some help!

This is the command I used.

aticonfig --od-setclocks=840,1250 --adapter=0
aticonfig --od-setclocks=950,1375 --adapter=1
aticonfig --od-setclocks=950,1375 --adapter=2

[supa]

It's easier to go with

aticonfig --odsc XXX,XXX

Your memory clocks seem awfully high - you can't set them any lower?

[minerX]

It's easier to go with

aticonfig --odsc XXX,XXX

Your memory clocks seem awfully high - you can't set them any lower?

No.  They are at the lowest setting.  I think I will try the RBE method next to lower the memory.  But I'll want to overvolt, overclock, and lower the memory.  0 is 6950, 1,2 is 6970

Is here my -ODGC

:~$ aticonfig --odgc --adapter=all

Adapter 0 - AMD Radeon HD 6900 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    800           1250
             Current Peak :    800           1250
  Configurable Peak Range : [500-840]     [1250-1325]
                 GPU load :    97%

Adapter 1 - AMD Radeon HD 6900 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    880           1375
             Current Peak :    880           1375
  Configurable Peak Range : [500-950]     [1375-1450]
                 GPU load :    97%

Adapter 2 - AMD Radeon HD 6900 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    880           1375
             Current Peak :    880           1375
  Configurable Peak Range : [500-950]     [1375-1450]
                 GPU load :    96%
:~$

[supa]

Are you using AMDOverDriveCtrl and leaving it running?

On the 58xx series, it drops the memory clock floor to 150 (at least, it does on 5850s and 5870s that I've had).

[minerX]

Are you using AMDOverDriveCtrl and leaving it running?

On the 58xx series, it drops the memory clock floor to 150 (at least, it does on 5850s and 5870s that I've had).

No.   I have tried twice on brand new installs with the same results.  I added no additional programs.   I haven't tried messing with the memory as I want to solve the overclock issue first.

[supa]

I say..... Go get AMDOverDriveCtrl and see if you can back off your memory clocks.  It will make a definite difference if it doesn't clear up your problem.

[minerX]

I say..... Go get AMDOverDriveCtrl and see if you can back off your memory clocks.  It will make a definite difference if it doesn't clear up your problem.

Can you point me to any command line instructions?  All I could find was the GUI that could only control the first card.

[supa]

I think -i X let's you choose adapter X.

[SchizophrenicX]

go to the terminal (once AMDOverdriveCtrl is installed) and type

AMDOverdriveCtrl -i0 <- for adapter 0
AMDOverdriveCtrl -ix <- for adapter x

AMDOverdriveCtrl is case-sensitive. You can get the index from the terminal output of any of them, use the index of those that are marked 'active'

[minerX]

Ok,

So I actually used RBE to get the clocks I wanted.   However the same problem has occured.   POCLBM is showing a LOWER hash speed when I overclock.


Stock speed @ 880 is 370Mhash/S.    Overclocked @ 1000 I'm getting 340 Mhash/S.


Adapter 0 - AMD Radeon HD 6900 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    250           150
             Current Peak :    800           1250
  Configurable Peak Range : [500-840]     [1250-1325]
                 GPU load :    4%

Adapter 1 - AMD Radeon HD 6900 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    1000           350
             Current Peak :    1000           350
  Configurable Peak Range : [500-1050]     [350-350]
                 GPU load :    97%

Adapter 2 - AMD Radeon HD 6900 Series
                            Core (MHz)    Memory (MHz)
           Current Clocks :    250           150
             Current Peak :    880           1375
  Configurable Peak Range : [500-950]     [1375-1450]
                 GPU load :    0%


11/06/2011 22:31:54, 4950c561, accepted
11/06/2011 22:31:55, dfce6ff3, accepted
11/06/2011 22:32:23, f4282659, accepted
11/06/2011 22:32:29, d549300e, accepted
11/06/2011 22:32:50, a9e0d574, accepted
11/06/2011 22:32:52, 224ca746, accepted
11/06/2011 22:32:56, 106f30ef, accepted
11/06/2011 22:32:57, 63a8156e, accepted
11/06/2011 22:32:59, 3f162021, accepted
11/06/2011 22:33:00, b28d6050, accepted
11/06/2011 22:33:13, 69b7c7a0, accepted
11/06/2011 22:33:23, 564095f0, accepted
11/06/2011 22:33:39, 8c67ee99, accepted
11/06/2011 22:33:48, d7213f2c, accepted
11/06/2011 22:34:33, deb8d541, accepted
338660 khash/s



Is it possible this is a miner error?

[dudel42]

go to the terminal (once AMDOverdriveCtrl is installed) and type

AMDOverdriveCtrl -i0 <- for adapter 0
AMDOverdriveCtrl -ix <- for adapter x

AMDOverdriveCtrl is case-sensitive. You can get the index from the terminal output of any of them, use the index of those that are marked 'active'

Is there a how to install AMDOverdriveCtrl? I tried building from source, but it wouldn't compile? binaries somewhere or docs on how to build yourself?

thanks.

[supa]

Alright, minerX!

What parameters are you using for poclbm?