ATI 5970 Artifacts / Corruption

[eth0s]

Every now and then this ATI 5970 ends up doing something like this: http://imageshack.us/clip/my-videos/837/o91.mp4/

It also does something similar (but not as bad), every time in 3DMark06. Games like Bad company 2 don't seem to set it off, apart from maybe a flicker every now and then.

To give you some background, I replaced the notoriously rattly fan a little while back. The issue looks memory related to me.

Things I've tried:

Changing drivers
Reinstalling Windows (had to anyway)
Flashing the vbios to the latest version from: http://forums.bit-tech.net/showthread.php?t=185918
Taking the card apart, reapplying thermal paste and giving it a blast of air

I was wondering if it has anything to do with the thermal pads, as they look a bit worse for wear: http://i.imgur.com/QASo1.jpg

To add, I can cold boot the machine, head straight to 3DMark06 and see the issue straight away which leads me to think it's not a heat issue. I can replace the thermal pads for around £16 it seems, but not sure if it'll be a waste of time.

Anyway, any other suggestions / tips welcome.

 

[americanbrian]

I agree it seems like a graphics memory issue. Did your original heatsink unit have a plate to cover the RAM chips (to extract the heat from them).

Best way to check if it is the ram is to seriously underclock it using a tuning tool, then stress test it.

Obviously this will temporarily negatively effect performance, but it is just to check if you resolve the problem.

I think some new high performance thermal pads on the RAM should do the trick TBH, but if it is not the RAM then it would be a waste of money.

 

[eth0s]

Thanks for the reply Brian. From what I can gather there's simply the TIM pads to make contact between the memory chips and the heatsink.

Good idea about massively underclocking the memory, I will give this a go tonight and update the thread.

 

[eth0s]

Hi, have you tested it with another power supply, monitor if you don't have one connect it to your tv, connecting your dvi-cable or hdmi cable to other dvi ports on the graphics card, pci-express power connectors properly connected? boot into safe-mode and see if still happens.

Could be your PSU, monitor or graphics card. Should not have taken the card apart and flashed the bios, that should have been the last resort. Test with another GFX card or onboard video if you have any.

I've tested it with separate monitors, cables and have reseated both the card and power cables.

I don't have another PSU to test it with currently, although I'm pretty confident it's memory related and not power related (famous last words).

In regards to taking the card apart, it's a pretty simple process and these cards are well known to rattle which was driving me crazy. Replacing the fan solved this.

Updating the vbios was worth a shot, the new version is known to fix some random bugs including artifacts, also I have a copy of the original vbios saved so no harm done trying.

I've had this card in two motherboards and it does the same in both.

 

[eth0s]

Unfortunately I don't have spare hardware to really test properly, for example to test the PSU with another graphics card I'd want to use one that'd put a similar load...

I do have a spare machine, but it's PSU is pretty weak so I doubt it'd cope- might be worth a shot.

I've had the card for about a year I think and haven't really bothered overclocking it- I dabbled one night but that was it.

First instance I'll try and underclock the memory and see if it makes a difference.

 

[eth0s]

Okay, so- I reduced the memory clock to 900Mhz (with AMD GPU Clock Tool) earlier and this has done the trick. The CC Overdrive Gui only goes as low as 1000Mhz.

The main issue with using AMD GPU Clock Tool is that it doesn't underclock the GPU when idle, the 5970 normally sits at 400Mhz or so, when using the tool it was setting it to 725Mhz constantly.

To get around this I used RBE: http://www.techpowerup.com/rbe/ to change the default vbios settings down to 900MHz and flashed the card again.

The underclock doesn't cause a massive loss in performance, in fact here's a comparison in 3DMark06: http://3dmark.com/compare/3dm06/16007897/3dm06/16010151

I'm fairly confident this could be resolved with new thermal pads, but I'm happy with this workaround for 5 minutes... thanks for the tip Brian.

 

[americanbrian]

You're very welcome. I don't suppose you fancy selecting my post as best answer? I am building up my reputation slowly, and would really appreciate it.

 

[americanbrian]

Wow, considering I have been an active member for many years more than you that seems a bit rich.

I am only trying to help, and it costs nothing for the OP to use the existing forum system to acknowledge a helpful answer.

Have a nice day.

 

[eth0s]

Well, BF3 seems to set this off- getting corruption on the outside BF3 Beta map and I've even tried clocking the memory down to 750MHz.

The only thing I haven't tried is underclocking the memory (tried clocking it up at bit), I can't see to find an application that can do this.

Was thinking about replacing the thermal pads to see if this makes any difference, but not entirely sure what thickness to use with the default heatsink, 3mm which can compress down to 1.5mm "seems" about right?

Was looking at these: http://www.watercoolinguk.co.uk/p/17mm-x-100mm-x-3mm-TIM-Pads-2-Pack_639.html

Opinions/thoughts?

 

[americanbrian]

For the price I would say give it a shot. You have a pretty sweet GPU there that is kind of useless if you can't actually run the games you want without it corrupting.

I hope that they fix you up good. Let me know how it goes...