Hardware Help - 5970

[Al]

Okay, not sure what happened, but here's my story. My 5970 was running MW pretty well, then I noticed a few errors. Then the errors increased to about 1 in 20 wus. I rebooted, no improvement. I then reinstalled the current driver, 12.10. No improvement. Uninstalled, DriverSwept and cc cleaned. Installed 13.1. All mw errored out immediately and I kept seeing the "ati driver has stopped unexpectedly and recovered" message. Reinstalled the 13.1, no change. Uninstalled, Driverswept and CC'ed. Reinstalled 12.10. MW would run on one gpu to 20% and then stall. The other gpu wouldn't start at all. Uninstalled and reinstalled again. Same thing. Tried Collatz...the wus wouldn't even start, but did give me the "ati stopped..." message. Tried WCG and the 5970 runs 2 wus on each gpus without a problem. Tried Albert and it runs 2 on each gpu without a problem. So, less gpu intensive projects run okay. The ones that hit the gpu hard, are failing. I'm assuming something died on the card. Any ideas would be appreciated.

[Fire$torm]

1st guess is the card's interior needs to be cleaned out.

2nd guess is that the TIM on the GPU has dried out. Note: Very common with ATI.

3rd guess..... both of the above.

[kmanley57]

I concur with number 2, MW runs my ATI harder than the others does, so it has to draw more heat off. But Number one needs done also, to blow off all the BAD bits!

[Al]

1st guess is the card's interior needs to be cleaned out.

2nd guess is that the TIM on the GPU has dried out. Note: Very common with ATI.

3rd guess..... both of the above.

I replaced the fan several months ago and the dried out Thermal Paste was replaced with AS5, so that should be good. I'll check for dust bunnies, but I don't think heat is really the issue at this point. Back when Hydra was making it's final push (a couple of weeks ago) to 100 million, this 5970 was picking 'em up and knocking 'em down on MW. Temps never got above 69c. I'll try stressing it with Furmark and see if I see any problems, especially heat.

Edit: I haven't been able to stress it with Furmark because it keeps defaulting to testing the 570 and I don't see a way to force it to test the ati. Maybe I need to remove the 570.

As an alternative, I took the core clock down to 500 and tried mw again. It started crunching. I'd let it run a couple of wus then increase the clock. I've taken it up to 675 (stock on a 5970 is 725) and I'll leave it there overnight. Since I rarely oc this card, 675 vs 725 is not a deal killer...about 5 seconds slower. Also, cutting the memory clock all the way down and the downclocked core clock keeps the temps at 60c with 60% fan. So far, no errors.

[STE\/E]

I know a lot of the 5870's I had would lose the ability to overvolt them, if I did the card wouldn't run right. As long as I ran them with a slight Overclock with no Voltage increase they ran okay though. I don't know if the same problem could occur in the 5970's ???

[Al]

I know a lot of the 5870's I had would lose the ability to overvolt them, if I did the card wouldn't run right. As long as I ran them with a slight Overclock with no Voltage increase they ran okay though. I don't know if the same problem could occur in the 5970's ???

The 5970 is a really poor design. My understanding is that the 2nd vrm is cooled by the heated air used to cool the 1st vrm, so the temp of the 2nd really controls how much you can oc or overvolt. In theory both gpus will run at the same clocks and volts as a 5870, in practice...not so much.

Not sure what is up with card, it may just be dying. It did run mw all night at the decreased clocks and only produced 1 error and it produced 135,000 credits, so it's still usable.

[Bryan]

Worse than that Al, the 2nd VRMs aren't under the cooling cavity at all ... very very poor design.

You need to check this out on MW ... call up GPU-Z and look at the temps of the VRMs. They should be kept below 110C. Each of the GPUs will go into self protect mode at 120C.

2nd there is a failure mode that you can see by looking at the current being drawn on 1 of the supplies. It will be way high in comparison to a good card ... I don't remember the numbers but I'm thinking it is similar to a 5870 for total current on a good card. It is on something like a 1.2V supply.

I never had a problem with the a 5970 not starting as you are. What I would see is that MW and Collatz would start giving me errored WU from 1 of the GPUs. Moo BTW is dumb enough that it will continue to give max credit even with one of the GPUs throwing errors. That tells you how valid that project is

If 1 GPU actually craps out and you can get it running stable on 1 GPU you can run it like a 5870 if it is the one with the good heat sinking capability. You can overvolt it and run it up to 900MHz+. You can use the "skip" command (IIRC) in your config sys file and just have it use the good GPU.

[Fire$torm]

The 5970 is a really poor design. My understanding is that the 2nd vrm is cooled by the heated air used to cool the 1st vrm, so the temp of the 2nd really controls how much you can oc or overvolt. In theory both gpus will run at the same clocks and volts as a 5870, in practice...not so much.

Not sure what is up with card, it may just be dying. It did run mw all night at the decreased clocks and only produced 1 error and it produced 135,000 credits, so it's still usable.

If the screen isn't showing any pixelation, which would be a sign of overcooked VRAM, then it might be the Vregs that are getting cooked. That is another common problem with ATI GPUs. Some of the 5xxx cards report Vreg temps and can be monitored with GPU-Z (Link).

[conf]

Any members still running this card ?
Just found one for 199,- Euro and it uses less power than the Asus card with 2 760s.
295W is acceptable but whats about the RAC earned from this amount of power ?

[Bryan]

Any members still running this card ?
Just found one for 199,- Euro and it uses less power than the Asus card with 2 760s.
295W is acceptable but whats about the RAC earned from this amount of power ?

I used to have 2 of them and they are VERY POORLY DESIGNED cards. The VRMs for the back GPU are NOT under the heat sink chamber so they run extremely hot. There is no way that I would purchase those cards. The cards I had were under lifetime warranty and each card was replaced 3 times in a 3 year period.

The 6990 which I also had was a great card. They learned from the 5970 and moved all VRMs under the heat chamber and totally solved the heat problem.