Random GPU going down and restarting days later?

[bboques]

I have 18 Cores mining for me and almost every day a different one goes down and then, if not touched, goes back up later.

XX@gmail.com_B2-1   XX 319957 (0.81%)   Prop.
XX@gmail.com_B2-2   XX 294898 (0.84%)   Prop.
XX@gmail.com_B2-3   XX 262899 (0.91%)   Prop.
XX@gmail.com_B2-4   XX   310562 (0.85%)   Prop.
XX@gmail.com_B3-1   XX   69755 (1.17%)   Prop.

If I didn't touch this it would come back hours later, and likely another would go down.

I'm Using Linux Ubunutu 11.06 on all machines.  Some are ADM32s others are Intells 2100s, my parameters are usuall worksize=258, aggresion=8, askrate=10, BFI_INT, Vectors and I'm mining with Pheonix.


Any ideas?

[gigabytecoin]

I experienced the same thing.

Honestly the only solution I found was to keep the core clocks at only a 5% overclock increase instead of the 10-15% many people report possible on these forums.

Good luck.

[bboques]

thanks for the feedback, I do not overclock since they keep going down.  Maybe I have to underclock! And with watercooling they are 45C so it makes no sense.  only 1 out of 4 of the cards on each of my 4 rigs was up today.  But if I don't mess with them, in a day or so they will go back up, and others will fall I don't get it.

[RyNinDaCleM]

Do you have full cover blocks? If not, it could be VRMs overheating.

Try a different miner! Can you use CGminer on linux?
See if that helps.

[phorensic]

Do you have full cover blocks? If not, it could be VRMs overheating.

I'm going to agree with this train of thought.  After months of mining and overclocking on my 5830's I've learned a few things.  First - automatic fan control will not work at all.  I have to set my fans at fixed 65% or more.  This doesn't just keep the core cool, it cools off the entire board (read: VRMs).  If I go lower than 50% fan my GPU's will freeze even though the core temperature is reported to be in the 60c range.  This amplified by the fact that I am open-case.  I don't have airflow over the mobo like inside a case.  If you waterblock doesn't hit all the areas on your GPU's and you are open-case...I wouldn't be surprised on the lock ups.

[bboques]

thank you guys, never heard of VRMs but Ill look into it