Topic: BAMT version 0.5 - Easy USB based mining Linux with farm wide management tools - page 52. (Read 324169 times)

Side topic comment: I leave my fans at 86% fixed, haven't had the same trouble as some others with the 5830 Sapphire (BitcoinXtreme versions lol). Small sample size but judging from what I've read from datacenter studies, Google has published several, reliability isn't that different between low and high for most computer parts. Main exception if I remember right was hard drives which both Google and Microsoft have found are the least resilient to high temperature and constant use.

The quality of the fans in various samples probably has more to do with fan life than the way that you use them.  However, I do have several models from several manufacturers that have run for nearly 1 year at 100% without any issue.

No... but 3 5870s at 100% all failed in 2 1.5 months, where 6 5870s with autofan have yet to fail in more than 4 months. I should have a larger set to choose from, true.

I suppose you did an exhaustive study with a large control group of fans not set to 100, compared to a large set of samples with fan set to 100?
Because I did..  and found the fans most likely to die were those on "auto" where they varied in speed.  In fact none of the fans set to 100, 85 or 50 had any failures at all.

My aching and dying fans have told me otherwise in the past.

No.

Running fan at 100% is fine.  Changing its speed a lot can in fact lead to premature failure.  Best to set it at some large value and leave alone.

No don't wait a week for the 7990's to come out.. much more worth the money. 1.3Ghash per card cost 849.99

Thats why I am selling of a lot of my 5 series and 7970's

Holyshit! Good to know.

* gigavps starts plotting a purchase of 7970s.......

Thanks I will try that later!

They are not supported "out of the box" in 0.5c, but I got 7970's working by doing the following:

Starting with a fresh install of bamt 0.5b (later versions should work i would assume).

• X does not start on first boot.  Have to manually type startx and it loads desktop.  Desktop says no usable gpus are detected.
  I start a root terminal and change root password.
  
  
• Remove previous version of flgrx (from https://wiki.ubuntu.com/X/Troubleshooting/FglrxInteferesWithRadeonDriver)
  sudo /usr/share/ati/fglrx-uninstall.sh --force
  coldreboot
  
  
• download new 7970 drivers:
  https://bitcointalksearch.org/topic/7970-linux-drivers-released-61987
  
  
• chmod +x amd-driver....run file to make it executable
  
  
• execute the driver install file
  /home/user/Downloads/amd-driver....run
  file to install drivers.
  
  
• coldreboot
  
  
• 7900 series detected... start mining with cgminer.

EDIT:  One person I was trying to assist could not get his 7970's to work, even following my method above... However I have done several fresh installs (to verify) and its worked every time for me.  I have 2 rigs running 7970's only.  They each have 3 x 7970's and work just fine for me on BAMT with this method.

You are thanking him for useless and incorrect advice? According to your posts and config, you were not overvolting or underclocking as he suggested and he told you to change a commented line while leaving it commented....hmph. 

BTW, no offense Isokivi, you did warn him about your knowledge level. I fully support your efforts to try and help him.

Are 7970's supported in 0.5c?

Aggressive way:
I run mine and make sure they are stable at stock gpu speeds with mem underclock for a day or two.  After that I try running at 800 MHz gpu clock.  Out of 6 cards (12 cores) only two cores could not run 800 MHz (they run at 790 and 795 MHz just fine).  My other 9 cores run between 800 and 855 at stock voltage... I have one "magic core" that is stable at 880MHz at stock voltage.  If 800 MHz fails during the next day, i decrease that core(s) by 10 MHz, if it succeeds, I increase by 25 MHz, until they fail, then decrease by 10 MHz until they run stable.  I recommend you do it while you have the ability to reboot the machine to minimize downtime.  You can get pretty close to stable OC's in a few days with the only downtime being a few minutes while you reboot (if/when needed).

Safe way:
Increase clocks by 10 MHz and let run for a day... then repeat.  Takes longer to find stable OC's, but allows for higher uptime, especially if you are not going to be near the rigs.  I did all my rigs the aggressive way, but had to revert to the safe way on one of my rigs as one cards OC was causing others to lock and I would change the OC on the wrong card... isolated issue from what I hear, but the safe way helps eliminate issues like that.

I had stability issues with my mem at 175 on my 5970's, so I run at mem of 190... something to consider... I got all my OC's stable at 310 mem, then ran 190 for about a week... then tried 175 and had issues, so I moved it back to 190... not sure if thats just card specific or what... but I didn't spend much time on it as the difference is negligible at that low of speeds anyways.

Also, earlier in this thread is a post by lodcrappo about OC'ing eating into your profits, you might want to go read that as he has some numbers about how much ROI you get for overclocking and hours of downtime don't make up for a small overclock increase... Its a good thing to read and consider while you are looking for stable clocks.

I run everything with air cooling and no overvolting, so YMMV depending on your setup.

Thanks for the tip Isokivi! I've deleted the file /live/image/BAMT/CONTROL/ACTIVE/noOCx and my Mem is now back to normal 175 Mhz speed.  Other than that, are there some clocks I should try to get my performance up with minimal compromises?

Do a search in this thread for "Mother" you will see why your card is at 725 and 1000.

Also, you may want to reconsider 100% fan...unless you enjoy replacing fans every few months.

Finally, about the comment above....the # in front of a line means that line is ignored so....# core_voltage_2: 1.125000 would not change the core voltage. It would do nothing until the # is removed.

Thanks guys - here is my corresponding BAMT dashboard: