7900 GTX cards keep dying

HeavyH20

[H]ard|Gawd
Joined
Oct 28, 2004
Messages
1,180
Here is the deal. I have two SLI systems. One is air and one is watercooled. That is four GTX cards, of which all 5 have failed (even the RMA replacement). The core was strong on all the cards (coolbits detect 700 plus) even after they started to fail. It is the memory or a memory control device that is at issue not the crap about vertex deltas and cores not handling load at pcperspective.

Two were EVGA GTX EGS and two were GTX SC cards. Different machines, PSUs, motherboards and CPUs. They both had the same issues. After about 30 days, the memory starts to go south, slowly but surely the memory needs to be downclocked further to avoid artifacts in games. Currently one of the SC cards is down to 650 (where it always seems to stabilize) and the second card is now down to 700. I cannot get past the last test in 3DMark06 without a lock up on either machine at anything above this. And, setting the core lower or higher does not affect stability. Just the memory.

There are a lot of people with these cards, and they are not all complaining, but it seems that the degradation occurs when running certain games, like BF2. Most other games work fine.

There is definitely something up with the 7900 GTX cards. I can't speak to the GT cards since I did not experience those myself, but it sounds like there is a similar problem. XFX has gone as far as shutting down their forum to avoid all the complaining and from one account, indicate they will not RMA the card since ALL the 7900 series NVIDIA cards have a similar issue.

I am disappointed as I really like these cards, when they work. I think I will cut my losses and return the SC cards since they are well within 30 days of purchase. I will have to keep the other two cards, put a single in each, and hope NVIDIA figures out their issues. I tried to give them some slack since the G71 is a new chip, as it is not a superset of the G70 and sports an all new core. But, crap is crap, and no matter how much you polish it, you still got crap.

Giving up on NVIDIA. Maybe there is a nice Crossfire system in my near future. :D
 
uhh well if your clocking SAMSUNGS 1.1ns memory higher then 900MHz, it should be failing as it cant handle that, not matter the cooling....
 
That's bizarre and the first I've read anywhere about that many massive failures period, and am boggled that there could be 5 from a single user. There was the vertex clock issue at release, but damn man, that sucks. I guess I've been really lucky, and yes they are touchy beasts, but my BFGs have been and continue to be awesome. I've run them hard since the day the GTX was released.
 
Opie said:
That's bizarre and the first I've read anywhere about that many massive failures period, and am boggled that there could be 5 from a single user. There was the vertex clock issue at release, but damn man, that sucks. I guess I've been really lucky, and yes they are touchy beasts, but my BFGs have been and continue to be awesome. I've run them hard since the day the GTX was released.

yea, bfg did alot to the memory, and totally messed the bios of that card, including changing the timings. (which is why it outperforms the 7900gtx superclock from EVGA) and bfg has absolutly wicked technical support. so i think im going to go with bfg for my next upgrade, just because of tech support and this review...
 
Yeah thats news to me also. I snagged a pair of EVGA 7900GTX EGS the day they came out and have nothing but success with them. I run them both OC'ed to 700 / 1780 without problems. Weird.
 
Well I havn't had any issues with my XFX 7900Gts but thats probably cause thier gts and arn't affected by the gtx issue.
 
MrWizard6600 said:
uhh well if your clocking SAMSUNGS 1.1ns memory higher then 900MHz, it should be failing as it cant handle that, not matter the cooling....

Funny thing is, the GTs, which have 1.4 ns mem, can overclock to 900+ MHz as well. I never did understand the reason for that. Wouldn't 1.4 ns mem have lower timings also?
 
The_Dark_Abyss said:
Funny thing is, the GTs, which have 1.4 ns mem, can overclock to 900+ MHz as well. I never did understand the reason for that. Wouldn't 1.4 ns mem have lower timings also?
so you are saying that people are getting over 1800 out of 1.4 ns memory?? i thought they were doing good to get 1600.
 
I have 2 BFG 7900 GTXs, purchased about 5 days ago. Both have similar problems.
Ran strong out of the box. 3rd day, system seems stable, start doing some benching
SM3.0 marks go to hell in 3DMark2006. Seems to either locked completely or for about 10 secs during Deep freeze. 3dM2003, 2005 crash. Thought it may be the drivers, tried a couple different versions, all the same. Thought it might be the power supply, put the cards on a seperate PSU, no change.
3dmark 2001 locks up during point sprite test. Both cards exhibit same problems be it in SLI or single card mode. Some Games run alright, artifacting after a time. I replaced with original 7800GTXs and system runs like a champ.
 
The_Dark_Abyss said:
Funny thing is, the GTs, which have 1.4 ns mem, can overclock to 900+ MHz as well. I never did understand the reason for that. Wouldn't 1.4 ns mem have lower timings also?

yea thats... new to me...

huh... so my dreams of 7900gtx oced to 725 have been pissed on? maybe a driver can fix this.... donno how but... it could.....
 
GTs have 1.4 ns memory, I thought they had same memory as 7900GTX since my 7900GT can also hit 570/900 on stock cooling (w/o vmods)...
 
There's a reason ram is rated for a certain __ns. And there's also a reason nvidia sets the speeds for different chips at different speeds.(max reliable long term speed) Granted there are always variations, but if you're overclocking then you shouldn't be complaining about your card dying.

Almost all of the videocard vendors are now doing a little more testing (speed binning) and selling overclocked versions. So if you buy a card from eVGA (who speed bins their cards) you should only expect what its clocked at, b/c if it ran faster they'd would have sold it as the top speed notch. And even that top speed notch is so overclocked, that you shouldn't necessarily expect much more.
 
You obviously have money so I'd be happy to sell you my guaranteed functional XFX Extreme Edition 690/1750@standard gtx. For a small premium of course

:D

Are you sure its not something silly like your PSU dying, seems odd that they all would have died.
 
Just got my eVGA 7900GTX EGS last week and have been playing all kinds of games at maxed out settings. Condemned, GR:AW, BF2.....and not a single problem....yet. :cool:
 
Iratus said:
Are you sure its not something silly like your PSU dying, seems odd that they all would have died.

I agree...the chances of five cards just happening to die on you doesn't make sense at all. I'd question whether the problem is on your end.
 
I've had 3 EVGA 7900 GT SC cards die on me in almost the same manner. The cards work great the first 2 days or so, but memory issues will eventually crop up, and I end up having to underclock the RAM by 100-150mhz to get things stable again.

The Deep Freeze test on 3dmark 06 will almost always lock up my system once the issues start occuring, so yea, the original poster isn't the only one having issues. I suspect there are many other 7900 cards out there that have started going south, people just don't realize it cause the issues don't appear in all games.
 
I am on my second SC and its artifacting , i am not even overlcocknig. WTF , It works fine with TRAA off but thats bullshit. No issue if i dont run nairtive resolution but thats bull shit.

I want a free G80 for my aggrevation
 
rk said:
I've had 3 EVGA 7900 GT SC cards die on me in almost the same manner. The cards work great the first 2 days or so, but memory issues will eventually crop up, and I end up having to underclock the RAM by 100-150mhz to get things stable again.

The Deep Freeze test on 3dmark 06 will almost always lock up my system once the issues start occuring, so yea, the original poster isn't the only one having issues. I suspect there are many other 7900 cards out there that have started going south, people just don't realize it cause the issues don't appear in all games.
what psu
 
BFG 7900 GTX here. Over 3 weeks of hard use and not a problem. I have it running as it came out of the box. Nice IQ and frames at 16x12 in GRAW,FEAR,HL2:LC, etc. I dont play BF2 however. My son uses it for Oblivion, he says it runs like a dream.
 
chrisf6969 said:
There's a reason ram is rated for a certain __ns. And there's also a reason nvidia sets the speeds for different chips at different speeds.(max reliable long term speed) Granted there are always variations, but if you're overclocking then you shouldn't be complaining about your card dying.

Almost all of the videocard vendors are now doing a little more testing (speed binning) and selling overclocked versions. So if you buy a card from eVGA (who speed bins their cards) you should only expect what its clocked at, b/c if it ran faster they'd would have sold it as the top speed notch. And even that top speed notch is so overclocked, that you shouldn't necessarily expect much more.

which isn't always the case either, most of the various 7900GT models, from the base GT to the Superclocked models are reaching the same top clock speeds roughly, i know my 7800GT N518 base model GT clocked to 530/1260 right out of the box

perhaps it was a KO Superclocked model that ended up getting flashed to a base GT model BIOS to fill orders, maybe not, but i haven't seen alot of results that say the higher end models are able to overclock further, or that they reach a higher ending clock speed in the end
 
Could it be temperature related? What kind of cooling do you guys use, both on the video card and in the rest of your case? Water-cooling a GPU obviously has an impact on the RAM...
 
if you guys have a GT and are experiencing memory issues, look at this

http://www.evga.com/community/messageboard/topic.asp?TOPIC_ID=14558

you've probably seen it by now if you are staying current of the "state" of the 7900GT's, but it's definately worth applying sinks to these chips, cant hurt, might help

unfortunately, it seems that alot of users whose cards works fine and then developed problems, who then sinked these chips, still had problems, as if the damage is permanent, not really enough info to support the theory that if they were sinked to begin with maybe they would not have suffered the damage to begin with

i know my Superclocked GT will get sinked before i ever install it

as far as the GTX's go, i am sure they have the same chips, but in a different location

also, for you guys having memory issues, write down the numbers on the top of the ram chips and go to Samsung's web site and see which chips you have, they make GDDR3 modules meant to run 1.8v and 2.0v, the GT's run 2.0v to the memory, if you have 1.8v memory running 2.0v that could also explain a gradual degredation in stability
 
lopoetve said:

I had a 600W FSP Group PSU on the first 2 cards, and tested it out on the 500W Seasonic SL12 I have sitting around on the 3rd. Didn't make a difference.


nobody_here said:

Yea I saw that post also, and sure enough those two chips were running *HOT* while gaming. When I received my second RMA (third card), I chopped up an aluminum chipset cpu I had and slapped it on there with some frag tape (need two sinks btw, since the chips are different heights), but it didn't help. The heatsinks I added were still getting *really* warm, and it definitely had air going over it from one of my case fans. Ah well.

As for other cooling, I went with a Zalman VF-900, but the GPU doesn't seem to be the issue here. On a whim, I tried pushing one of the cards as high as it would go after the artifacting began. GPU hit 575mhz, but RAM would only be stable at 1200mhz.
 
EVGA tech support tells me its an issue with voltage regulatores overheating and not memroy or core. They are working the issue with Nvidia and it affects all factory oced cards from all manufacturers.
 
jacuzz1 said:
EVGA tech support tells me its an issue with voltage regulatores overheating and not memroy or core. They are working the issue with Nvidia and it affects all factory oced cards from all manufacturers.

yeah, that's the thread i linked to above, those are the chips that make the circuit that supplies the voltage to the memory and regulates it

originally it was thought that it was only a problem for voltmodded cards, but stock cards are failing too, like i said, i am sinking mine the day i get it
 
mavalpha said:
Could it be temperature related? What kind of cooling do you guys use, both on the video card and in the rest of your case? Water-cooling a GPU obviously has an impact on the RAM...

IT is but it is not. There are two chips on the bottom of the card ( at least the 7900gtx superclock )that are used for voltage regulations and these are what are overheating.
 
nobody_here said:
yeah, that's the thread i linked to above, those are the chips that make the circuit that supplies the voltage to the memory and regulates it

originally it was thought that it was only a problem for voltmodded cards, but stock cards are failing too, like i said, i am sinking mine the day i get it


LOL I forgot i already posted on the issue........... Yes i am am going to sink my second rma when i get it.

By the way , once you see the artifacts its too late......rma it
 
The issue only appears with HDR and SM3.0 based games. So, most people will never notice an issue. Once the first EGS failed, I decided to see if the rest of the cards could handle a few hours of SM3.0 based HDR games. They all failed. Once artifacts start, the cards degrade to a point where they are only stable at 650 MHz. They never degrade past this point, strangely enough.

The PSU's are a 600 Watt OCZ SLI PSU and a Fortron 700 Watt SLI PSU. Not that it matters, but the only delta between the system configs is the new video cards since I was running a 7800 GTX SLI and a 7800 GTX 512 SLI system prior. Just popped the new cards in and used them. The system configs, otherwise, remain the same. So, the sytems work with 7800 based cards, just not the 7900 series. And, using a micrometer, I verified the voltages are well within spec on the PSUs. I also use a SmartUPS UPS on each to ensure no power issues (brown outs, spikes, etc).

My theory, the memory voltage controller on these cards die under a cetain amount of use. SM 3.0 and HDR absolutely tortures the GDDR3 (you can hear it squeal). So, play BF2 for a bit or run the last test of 3Dmark06, and the card degrades. First the 800 stock spec worked on the SC. Then 750, then 700 and finally 650. The core is solid. No issues there. Play an older game at stock memory specs, no problems. Run a newer game, and artifacts, hiccups and lock-ups occur.

Just to make sure these were good cards, and would not die outside of my 30 days, I looped the last test of 3Dmark06 with HQ settings at stock speeds for 50 times. That is more HDR and SM 3.0 action than most people would put at their card in 3 months. So, the test speeds up time, and gives you a glimpse of the future. The cards will fail, not if, just when. They are nice cards up until they fail.

I wish I had my 7800 cards back.
 
i have two pieces of advice that everyone should follow:

1. make sure ALL the ram chips are well cooled. preferably no more than "warm to the touch" when stress testing
2. put heatsinks on all mosfets, and be sure that there is plenty of airflow over them
 
EVGA's overclocking has been disasterous since they started. Mostly all failures and problems that I have come across happened to be with them.
 
The Superclocked had a problem with some "delta" or something with the pixel engine or something. Don't know exactly what the specifics are, but because of the high clocks on the factory overclocked versions, something on the card overloads over time. That's why I'm buying two stock clocked 7900GTXs from eVGA....
 
HeavyH20 said:
My theory, the memory voltage controller on these cards die under a cetain amount of use. SM 3.0 and HDR absolutely tortures the GDDR3 (you can hear it squeal). So, play BF2 for a bot or run the last test of 3Dmark06, and the card degrades.

Yet BF2 doesn't use HDR :confused:
 
If that is true, then these cards are just BAD. So I'm running ONLY stock clocks on my cards. I have heard that these cards have almost zero headroom anyway due to the limitations of GDDR3...
 
Well they are going to do a 24 hour burn in on my next card since , from what i am told its not evey card that has the issue. I cannot wait to see if it make a difference. I have been told that 3dmak 06 for 24 hours and if it passes , they will ship it to me.

Time will tell.
 
Sovereign said:
If that is true, then these cards are just BAD. So I'm running ONLY stock clocks on my cards. I have heard that these cards have almost zero headroom anyway due to the limitations of GDDR3...

well, thats not really true, there are many people who are successfully running clocks like 700/1800 when stock was 500/1500 and still running them this way, the problem doesn't affect all cards
 
How would I tell if my card is "good" (not going to die from OCing) or "bad" (any tampering will kill the card)?
 
J-Mag said:
Yet BF2 doesn't use HDR :confused:

No HDR but I believe SM 3.0 is used.

As for the headroom, it is true that the SC versions do not clock any better than the regular cards. I verified this. But, the memory is the real distinction. The SC cards still coolbits detect around 930 or so while the regular GTX cards do 870 or so. Even though this detect still works, an SM 3.0 app will still have issues at the 800 MHz clock. It is a weird behavior.

The cores on all the cards do around 700 to 720 with the 50 Mhz delta in place. The cores never degraded. And, one of the regular EGS cards is the best clocker at 729. The SC cards place 2nd and 3rd with the second EGS bringing up the rear.

If you really want to test your card, simply loop the last test of 3Dmark06 exclusively for 50 loops. Does not take long, only about 2 hours.

And, for those wondering about cooling on the SC cards; the core, memory and power modules were all watercooled. An 80 mm fan was also blowing air across the back of each card. The GPU never went over 38C and the memory was also at 38C. The back of the card was cool to the touch at any time. So, even with more than adequate cooling, both SC cards failed the last test of 3Dmark06 at spec GTX clocks (650/800). Never mind their spec OC clocks (690/880).

The 50 MHz vertex core was always around, even on the 7800 GTX (OK, it was actually 40 MHz - only the ASUS TOP had the 50 MHz delta). There were plenty of those cards which had a 490 stock clock (540 net) when the stock spec was 425. So, this is really no different. And, again, I do not think the cores are at issue, it is some type of memory controller issue NVIDIA needs to remedy. If I simply clock back the memory, the cards all stabilize. The core on any of the cards could be at 650, 690 or 700 and never affect stability. Try raising the memory clock back above 650 on the issue cards, and the instability returns.
 
Back
Top