Topic: Gigabyte 7970 Ghz Edition is bad - What's good? (Read 11998 times)

Is there any BIOS for the card that can give you voltages between 0.95V and 1.2V? Seems quite a gap.

l also found that running at 700/0.95v knocks off over 100watts from the power consumption, obviously reducing the hash rate, but improving the rig efficiency/profit. Reduced temp from 76C to 56C,

Under-clocking ftw, though it would be nice to have firmware with a few more options between 0.95v - 1.20v

Even with the heavy under clock the card is still doing 500kH/s on scrypt, but the system is pulling 220watts vs 360watt when the 7970 was on defaults. at 24cents per kWh, that's significant.

At the current prices, something like 0.55btc each, you could buy 7+ USB erupter miners and a 3-4A d07 usb expander.  See current sale under "mining, group buys, "Canary in the Mine"

 Then run btgminer under w7 or linux.

Your USB would contribute to hash rate 7*333 >= 2200 mH. 

That is four times the hash rate for the price of one card.  And yes, you can run the GPU and the USB miners at the same time.

Your two GPU would add 1100 giving you 3300 mH.

I can get it done with MSI Afterburner, but it's unstable and requires much babysitting.  I just couldn't get it to stick, so I said F' it.

Does anybody managed to undervolt GV-R7970TO-3GD rev: 2.1 ?

The card has stock voltage locked at 1.256V.
I wonder if it is possible to decrease that number to let say 1.09V.

When engine/memory are underclocked the card allows to set 1.2v, but nothing less than that.

Yeah, if you can undervolt, it's well worth it. My 7990 (essentially 2x7970) does 950 MHz/700MHz at 1.00v. From the stock voltage of 1.17v that's a huge difference, brings temps down by 10C.

Ya I was just mapping the defaults out this past week, and here's what I got:

0.85V is anything 450MHz and below
0.90V is 451-600MHz
0.95V is 601-700MHz
1.20V is 701-1000MHz
1.256V is anything 1001MHz and above

I'm running it at 700/550/0.95V cuz I'm sick of the fan noise/heat in my bedroom, and it's almost the same profit (after deducting electric costs). Now obviously if I use MSI AB I can force a voltage, and find a higher clock stable at each setting. For example, my card is stable at 900MHz @ 950mV, which is 200MHz higher than the default 0.95V settings (or 0.25V lower than the default settings for 900MHz, depending on how you're looking at it).

One thing people might not have noticed about the GHz cards, is if you reduce the engine clock down to 1000Mhz, the firmware will reduce the core voltage down from 1.256v to 1.2volts. That may help some people with cooling problems.

I'm currently running two gigabyte 7970 Ghz cards.  I've seen it posted a few times that they will pull 1400 watts.  I ran mine for about two weeks with a 660 watt Seasonic power supply in a open case.  The top card would run hot and start thermal throttling at around 90 degrees.  The bottom card ran mid 80's.

I ended up buying a new case and power supply before over clocking the cards.  With a lot of tweaking I now have the top card running at 1160/750 and hashing an average of 700M/s at 78 degrees during the hottest part of the day.  At night it runs about 73-74 degrees.  The bottom card runs at 1210/750 and hashes an average of 730 to 740M/s while running 71 at the hottest part of the day..   At night the bottom card runs around 67-68. 

I've had this setup hashing 24/7 for more than a month now.  I lower the intensity on the top card when I play games, but it still does about 650M/s  while gaming.  I mine BBQcoins with the CPU for fun when I'm not using it.  Does about 30k/s without affecting my gpu's.  heh..

Afterburner won't let me turn the memory down any lower.  I've read if I close the program and reopen it that you can lower it  some more, but that hasn't worked for me.  How did you guys turn it down lower?

Chad

I use cgminer, which keeps the temps down, so I don't think the two of five cards of mine failed due to heat.  They are not GHZ versions.  I still have two more that are working just fine, and one that has to be babied to minimize HW errors.  Pretty bad stats:  2/5 toast.

+1

I also Have one of each, they need light airflow or yes they overheat.   I have a 180mm AP182, excellent fan and overkill but works well.

Sapphire OC 7970s @ 1090/1500 in Win8x64 12.6 drivers with sdk2.8 gets me 760's per card using scrypt.  They are also voltage unlocked.

I have 2 x Gigabyte 7970 Ghz Editions 1100/1500. These cards demand excellent airflow otherwise you'll have stupid temps as did I in the beginning 99C+ !!! They're a great card and clock very well despite being voltage locked :\

My setup: http://imgur.com/eWclj9s

My stats: http://imgur.com/ceKzyov

I have two 7970's - one GHz edition and one normal edition. Running them both on high I have no issue keeping temps under 80 degrees. I would highly recommend you look at getting some more airflow into your case. I actually modded my Bitfenix Shenobi to allow me to mount a 23cm fan directly blowing air onto the cards.

It is the GHz edition. Stock OC of 1100MHz?
I've heard of plenty of people using these cards with no issues. With the exception of typically low scrypt performance, they're not bad.

FWIW I am not a fanboy. I used to use XFX back when they had double-lifetime warranties. But for my mining rigs I went with two systems, each with 3 Gigabyte 7970s, the GV-R797OC-3GD model. I haven't tried to push them too far, but I have gotten about 700 kh/s scrypt, and 620 MH/s SHA256. They are voltage locked unfortunately, but I can OC memory for scrypt, and UC memory for SHA256 while OC'ing the core just fine.

They are in a milk carton open air case with 250mm fans blowing at them, and they run around 65C with 24C ambient temps.

Buddy 1 card - you may have been lucky and actually had one that works - Plus when you have multiple in one machine is when you will really see how bad they are.

The GHz edition is the better ($50 more) version. The cheaper version is the one that is voltage locked.

It's probably not the GHz edition then.

Come on now. I've had a Gigabyte running at 1200 MHz for over a year now. No major issues.

I'm now a believer.

I have bought 8 Gigabyte 7970 and 16 Sapphire 7970 and I can say with authority the Gigabyte cards are absolute SHITE honestly I will never buy another Gigabyte card in my life.

Some of them have to be under clocked to stop them falling over - one of them just overheats so much that it is sitting on my desk as I am unable to use it.

I can not emphasise enough how bad those cards are.

The Sapphire cards on the other hand are brilliant - overclock to 1175, hash around the 700mhs mark and remain totally stable.

Do not buy Gigabyte EVER

So sent the one giga 7970 for RMA, then another one in the same rig started doing it.  The second sparking 7970 is in a different pcie slot.  Both are on unpowered risers.  Sparks seem to be coming from different locations.  I hope to catch it on video sometime.  At least the PSU is shutting down....  At least.

PSU:  COOLER MASTER Silent Pro Gold Series RSC00-80GAD3-US 1200W
Mobo:  GIGABYTE GA-990FXA-UD3 AM3+

One of my gigabyte 7970s started sparking too!  Scary business.  There's no ozone smell of a good arc, but it's a pretty big burst of sparks coming out of the top heat sink.

Where can I get the standard firmware from? I would like to try 1.17v I am having trouble with temps (77-80C) even though I have opened my case up.

My 6970 runs with a memclock of 110 just fine; I underclock with "aticonfig --odsc=970,110". Trust me, I tried running my 7970s at 300, 500, etc.


The stuff I've been looking at is non-conductive, so I think it'll be fine.

Also, as another status update, I'm pretty sure that the hot GPU was defective because it shorted. Sparks were flying and everything. I'm honestly surprised (and fortunate) that nothing caught on fire. I've told Gigabytes and will hopefully get a refund.

As CK says of cgminer, the 7970 requires the memory clock to be no more than -150 than the engine clock.  Per the cgminer readme:

--gpu-memdiff -125
This setting will modify the memory speed whenever the GPU clock speed is
modified by --auto-gpu. In this example, it will set the memory speed to
be 125 Mhz lower than the GPU speed. This is useful for some cards like the
6970 which normally don't allow a bigger clock speed difference. The 6970 is
known to only allow -125, while the 7970 only allows -150.

I don't have any experience with the Nano Diamond. I've used AS Céramique, which is nice for GPUs cuz it's non-conductive, and pretty decent.

I have all but 4 of my fans plugged into molex, and I think there were 3 settings in the BIOS for fan speeds, which I set to "Full On", so at-worst there's 1 fan not running at max speed out of about 16. I'll buy some thermal paste after work and try reseating the GPU heatsinks when it comes in. I've heard that nano diamond is the way to go. Any experience on this?

Make sure that the fans on your case aren't being controlled by your motherboard. I have them plugged right into a molex adapter, so they're running at 100% off the 12V line.

Idk why your temps are so high. As I said, I'm OC'd to 1200MHz and I'm only using an Antec 300 case, and my temps right now are 69C. Try taking the heatsink off and repasting it? I always do that to any GPU I get.

Flipping the fans on the side did the trick. It's still running kind of hot for my taste (above 80C), but I'm not concerned about it anymore. Although this works, I still wouldn't recommend this card. I'll do a mountain of research once I have the time and see about getting a 3rd 7970, preferably without a locked voltage, I'll see how it compares and add to this thread.

I'm under-clocking with the the aticonfig command that came with the FGLRX driver. I use Linux, so my options are very limited *grumble grumble*


Interesting. I'll have to try this. Playing with my fan layout has seemed to yield the best results, so this will hopefully get it below 90C,


I was going to flash it with the non-Ghz edition BIOS. It's still voltage locked, but the voltage is lower, and the person who tried this said they were able to run at their original clocks (probably because the memory clock was so low) with lower temps.


I'm sure the source of my troubles is the locked voltage. I think a lot of my hate was the initial shock of all of the problems related to this, but now that I realize the cause I only have one major problem with this card. I don't get why the voltage needs to be locked in the first place; It's horrible for temps and power consumption, which are both very important factors in Bitcoin mining and gaming (though gamers probably don't care as much).

Ya 1180 isn't bad at all! What program are you using to underclock the memory? CGMiner won't work with the 1180/110 setting, but will think it does. Try using a 3rd party program like GPU-z and watch the sensors tab to see the actual mem speed, not the reported mem speed. Yes, they can be different. You also won't notice a huge difference with the 1100 setting. Try using MSI AB to set it to something around 375-410.

Ya mine actually ran cooler when I took the side of my Antec 300 off. I didn't like the dust, so I put it back on, and had the fan by the GPU blowing out, not in.

I would NOT recommend messing with the BIOS. I tried to flash a voltage unlocked BIOS, and now my card won't be recognized by any of the AMD drivers, and I can't mine. I had to shut down the computer, flip the Dual-BIOS switch, and it's been running on the backup read-only BIOS for ~6 months now, no issues. Even with this backup BIOS, it's still voltage-unlocked, but it just takes a LOT of mucking around in MSI AB to get it working.

I'm on Win8x64, 13.1 (and included SDK), CGMiner 2.11.0, E:1200  M:1050  F:90%  and ~72C for a constant 722MH/s per card. I've had great luck with this card! It is eventually possible to undervolt, but you gotta mess with it. At this point, it's more profitable to max your overclock, and not worry about undervolting, so that part kinda doesn't matter as much. The Triforce cooler keeps the Core Engine temp pretty low, but leaves a bit to be desired with the VRM cooling, especially with the higher-than-normal stock voltage, and the itty bitty heatsink on them. I run the fan at higher speeds for the VRM cooling, regardless of the Core temps.

All in all I love these cards, so idk why you would hate them so.

Do you have your driver updated to the most recent one? Do you use Windows? I've heard that for Linux with certain versions of FGLRX, dropping below the recommended clocks does weird things. Definitely try re-installing the driver,

Got a pair of 7970s running at core 1075 (stock 1000 Mhz) and memory 1200 (stock 1425), stock volts. Card is voltage-locked unfortunately. But I cannot get memory ro run at less than 1200. Any lower, and system crashes.

Any idea why? Others seem stable at much lower memory underclocks!

Sounds about right for those gigabyte's

I just checked my Steplight Powermeter and it reads ~1400W.

Yes, I've under-clocked my memory from 1500Mhz to 110Mhz. The problem is that this doesn't allow me to further OC my core clock, nor does it decrease power usage or temperatures.

EDIT: Update on my temps. After letting it run for a few hours, my temps are about 92C/78C Not sure what I can do to reduce these. I might have to resort to water cooling.

you do realize that for bitcoin mining, you're supposed to downclock memory A LOT, right? not sure how it is with 7000 series but I can downclock my 5850 from 1000 to 300 with hash rate increase!

Heheheh... If I ever have that kind of money laying around... Looks like it'd OC a little more than a 7990. Assuming I could fit 3 on my board...


Flashing my GPU BIOS is proving impossible without Windows. It'd be a shame as I've gone 2 years and 6 months without needing it. I'll look into the DIGI+ VRM and see what I can do. I have a good feeling about this.

UPDATE: after repositioning my fans and adding a few more, It seems it's stable at about 87C for top card and 78C  for middle card after about 30 minutes without outside air. This is still too high for my personal liking, but the system's not crashing, which is a start. I'll leave this running for about another 15-30 minutes while I research DIGI+ VRM and see if these current temps stay stable. If the DIGI+ VRM thing doesn't work, It'll be stuck at these temps for a while.

Well if you don't care about price, then buy this:
http://www.asus.com/ROG/ARES26GD5/

Dual 7970 GPU with hybrid air/liquid cooling. You can directly funnel the heat out of the case. Wicked expensive. But hey, you said "regardless of price" right?

And since you're persnickity:

Acclaimed DIGI+ VRM joins superior quality 20-phase Super Alloy Power for precise digital control, enhanced performance, reliability, and card longevity

GPU Tweak: modify clock speeds, voltages, fan performance and more, all via an intuitive interface

This is a grand idea. I'll do some research and see how this is done on Linux. I'll post back later with my results.

Power consumption from my Kill-A-Watt:

Idle: 167 W
SHA-256 hashing@1140/990, 1.17V: 417 W (~250W net), ~680 MH=2.72 MH/J
scrypt hashing@1080/1500, 1.17V: 438 W (~270W net), ~690KH=2.56 KH/J

Of course, YMMV.

Windforce is only cool if it can get fresh air

how much your watt consumation, with those?




really strange, i know for fact that the windforce 3 is very good in temp

Yeah I can either run at 1.170mv or 1.256mv but I cant undervolt 

I have 4 of them. http://imgur.com/a/mRbMZ#y6zZ4pi

Unfortunately I cannot undervolt them to save on electricity. Stuck at 1.256V once I go over 1000Mhz Engine.

QG

Edit: updated picture url

Depending on the manufacturer, you may be able to load different firmware on the card. My Gigabyte Windforce GHz edition 7970 came with a stock voltage of 1.25V; with their "standard" firmware (instead of "GHz"), that went down to 1.17V which has helped with temps and actually gets me better hash rate since I use auto core clock control in cgminer to maintain constant temp.

Oh, and try Florida if you think you are struggling with temps ;-)

I should have. I didn't do enough research before I bought these, so I didn't know it was basically 2 7970s put together. Maybe I'll just save up and get a 7990 instead of another 7970. My only concern is that the only 7990 I can find also has a locked voltage (which I think is source of my pain with the 7970s).


Yes.


That I do not. I have the cards running in this beast: http://www.newegg.com/Product/Product.aspx?Item=N82E16811147157

Are you sure temp is the issue causing the crash?
I highly doubt 90C would crash any GPU. I've got one running at 101C today, rock solid.

You need a new case.  I've got 2x of the same cards mining LTC (granted at like 1022 core 1244 mem for optimum khash) but have no issues keeping them below 80C and rock solid.

you talk about windforcce 3?

Then the sky is the limit. Or? maybe just get a few 7990 

Sadly, I'm in a small apartment right now. The lease is up in October, so when that happens I'll see about getting a place with a "server room". Letting it mine outside is also not a good idea because the weather in Alaska is crazy. Sunny to 80MPH winds and snow/rain after 20 minutes isn't surprising, even around this time of year.


I didn't realize 1180 was so good, even with the temps. My main reason for complaining is that under-clocking the memory, even to less than 10% of the stock clock, does absolutely nothing for the power usage and temperature. It's very frustrating, and I'm sure that other 7970s don't do this, while still getting the same hash rates.


At a first glance, it looks like the money/hash rate ratio is about the same, but I'd prefer higher hash rates regardless of price. More hash rate per cards means less machines, which means less money goes into cases, CPUs, RAM, PSU, Motherboard, and shipping and more goes into GPUs. The idea is viable, I'm just very persnickety and like to take things to the extreme.

Yes 7950s are a viable choice. Here in europe I compared the price and you getting excactly the same hashrate for your buck than with the 7970s.
I would definately open the case up, or put one of your case fans (if you have spare, like the ones on the top, or at hdd bay) to blow air right on to the cards.

Why not try 7950s? Although they have slightly lower Hashrates they're still quite cheaper than the 7970s.

You're getting 2 7970s inside a case to run at 1180 and you're complaining?

Be happy.  I have 2 of those same Gigabyte 7970s running at 1110/125 in an open rig running 85C/79C directly on the motherboard (ambient 80F-90F)

Open your case up and maybe put the miner in the garage if you have one, or build a cover and let it mine outside.

Unless you add a lot more core voltage you're not going above 1180 on most 7970s.

I left a review on NewEgg saying I would never recommend this card to anybody, and I'm here to take that a step further as well as ask what manufacturer has the best 7970 for BItCoins mining. My electricity is cheap, so power consumption is not an issue.

Anyway, this particular Gigabyte card is terrible:
 - Two of them in one case runs so hot that I need to let in cold, Alaskan air from outside to keep them cool. Crash at above 90C within 3 minutes at stock clocks.
 - Run stable at 1180/110 and 1180/1100, but not stable at 1190/110 or 1190/1100. All of these combinations have the same power draw and temperature, by the way.
 - I haven't tried water cooling, but based on how fast the temperature increases, I'm not convinced that'll keep them very cool.

Now the other question:
Which card has the highest hash rates? I sold off the Bitcoins I bought when the prices were low, and have just over $425 to put towards a 3rd card for my rig.
I've got an extra 550 watts on my PSU, so I don't think I have any real limitations other than temperature or money.