XFX 7970 Black DD nightmare

[cryptologist]

Quad Crossfire XFX 7970 DD Black Edition 3GB
ASRock 970 Extreme4
Athlon II X2 270 3.4GHz
8GB G.Skill Ripjaws DDR3-1600
64GB SanDisk SSD
OCZ ZX 1250w PSU
Win 7 Ult X64
CGMiner 3.2.1


I have a mining rig with 4/x XFX 7970 Black DD cards. They're brand new so unfortunately they're the neutered version without the BIOS switch and other cost saving measures secretly implemented by XFX without even changing the revision number. I wish I would have done my due diligence before purchasing them, but I didn't know any better at the time. They're also supposedly hard voltage locked. I can't change the voltage in Afterburner, CGMiner, Trixx, etc. even after enabling the unsupported core voltage setting.

I've built three other rigs, each with 4x Sapphire 7950's and all 12 cards are happily churning out 650 Khash/s mining Litecoin 24/7. These XFX 7970s are a whole different animal. I can't get them to mine Scrypt or SHA256 for more than a few minutes without crashing. It's not the typical driver crash where the screen goes blank for a few seconds and a single card shuts down, nor is it a BSOD. The screen will go black and never turn back on, but the computer stays running. The only way to recover is to manually turn the PSU off with the rocker switch. The 7950 rigs are all using the same RAM, SSD, motherboard, CPU, PSU, operating system, and CGMiner version as my 7970 rig and they are rock solid stable.

I've tried countless settings, including Catalyst 12.11b, 13.1, and 13.4 and ran the AMD driver cleanup utility each time. I've tried using powered risers, plugging directly into the motherboard, removing one card at a time, and switching between 1x and 16x powered risers to try different PCI-e configurations. I've tried running with just a value for shaders and I've tried all different thread concurrences. I've tried at low intensities and high intensities. I've tried 4 different brands of RAM. I've tried 1 thread or 2 per GPU.

Even at stock clocks, both Scrypt and SHA256 mining crashes after just a few minutes of mining. I think it may be related to temps, because the crash always seems to occur when one of the GPUs hits around 80 degrees. If I set CGMiner auto-GPU to a ridiculously low setting and have it disable each GPU at about 75 degrees and then turn them back on again, the system will stay running. The problem is, these GPUs get so hot so quickly that under this scenario, each GPU can only run for about 20-30 seconds before CGMiner disables them to cool off again. And this is with the fans manually set at 100%.

I know 7970s run hot and can hit 90 degrees under full load. What good are mine if they lock up at 80 degrees? I would love to be able to undervolt them because the stock setting of 1.175v seems like overkill to me for the stock clocks of 1000/1425, but I believe these cards are hardware locked so undervolting doesn't seem to be an option.

I've read through every 7970 and XFX-related thread on this forum and on the internet at large and I'm out of ideas. I can't find any evidence that it's possible to unlock the voltage on these cards and there don't seem to be any compatible BIOS for these cards besides stock.

If anyone has any insight as to why my system may be crashing at 80 degrees no matter what version drivers I'm running and regardless of CGMiner config settings, please let me know. I'm truly lost at this point.

[1714932213]

1714932213

[cryptologist]

I don't know why it took me so long to think of doing what I probably should have done from the beginning: use the "device" flag to disable one GPU at a time. It seem that the GPU that my display plugs into is the problematic card, which explains why the screen would blackout. After disabling this, the other 3 GPUs run fine well above 80 degrees. It seems I may have a bad GPU.

Any other insights would still be very much appreciated.

Does anyone know if there are any other BIOSes that will work for these cards or any way to enable voltage control?

[philips]

I don't know why it took me so long to think of doing what I probably should have done from the beginning: use the "device" flag to disable one GPU at a time. It seem that the GPU that my display plugs into is the problematic card, which explains why the screen would blackout. After disabling this, the other 3 GPUs run fine well above 80 degrees. It seems I may have a bad GPU.

Any other insights would still be very much appreciated.

Does anyone know if there are any other BIOSes that will work for these cards or any way to enable voltage control?

https://bitcointalk.org/index.php?topic=216474.msg2278380#msg2278380

You will need to force the flash.

[ssateneth]

This is a primary reason why I don't do non-ref cards. You never know what is missing until after you've already bought them. With reference cards, you know exactly what you are getting.

I'll trade you 2 of your non-ref 7970s for ONE of my pure reference 7970s