Gigabyte 280X Windforce Rev.2 rare artifacts

[ivicagmc]

10 days ago I got myself new 280x card. I get rare artyfacts, different type but most of the time are like on the picture... World ot tanks is were they are most often. I got them in Unigine bencmarks but just 2 times in like 30h of bencmarks, 2 times in crome after benchmarks. Crysis 3, Elite Dangerous, BF4, 3D mark nothing.
Now here is what happened so far.
Tried all possible change of BIOS. Even forced 7970 GHz edition BIOS and got even more artifacts. Change of drivers and stuff, nothing helps.
Underclock GPU 1050MHz, Memory 1400MHz got artifacts again in World of Tanks.
Start OCCT GPU error check on 2GB of ram. Stable.
Put memory on 1600MHz (100MHz overclock). System shutdown. I figure out that is PSU. My psu has 4 rails by 18A and only one rail goes for graphics. Not enough power from that rail so I put one 6 pin connector on molex and connect it to other rail for drives. No system shutdown but immediate errors in OCCT.
I put memory on 1510MHz errors after 5min.
I put memory on 1500MHz no errors after 30min, system stable.
Now here is weird part.
I put powerlimit -15%, memory on 1630MHz after 30min, system stable.
So it is defently not memory. Or am I wrong?
Now PSU is almost 3 years old and I cant afford new right now, spent money on graphics. Don't know anyone with enough powerful PSU around to borrow it.
I can return graphics but will they find anything wrong with it since common test don't show a thing and CAN I BE CERTAIN IT IS PSU AND NOT GRAPHICS? There is a bunch of stuff about defective R9 graphics cards and how people returned it and got same or new card but artifacts don't go away.
Thing is how come it is PSU when I get artifacts on mediocre PSU load and not any on full load?

 

[jboydgolfer]

sorry to hear , it'll get busy here soon. If those are Your +12V rails the PSU very well COULD be the problem...with age it gets worse.not to mention a 7970 eats Watts.

and no your shouldn't split the rails...I recommend NEVER to use those damn molex adapters, those are basically a way for the Card makers to get You to buy a new card sooner, by shitting your current one out.

 

[Ferrum Master]

Dissemble the card... see if the holder bracket isn't bent... the that holds up the heat pipe block.

Otherwise... welcome... your card is dying just as mine but... mine is already 2.5 years old...

 

[ivicagmc]

Dissemble the card... see if the holder bracket isn't bent... the that holds up the heat pipe block.

Otherwise... welcome... your card is dying just as mine but... mine is already 2.5 years old...

Already did that and put some Prolimatec thermal paste and lower temp for about 5 C. All memchips are well covered with thermal tapes, contact is good and screws are tight... So card isnt going to die, that is NICE thing to hear.
I found memtestCL compatible with this card. Did 50 iterations on 2GB memory and 0 errors. Here it is if anyone is interested.... https://github.com/ihaque/memtestCL

 

[Pill Monster]

Not PSU. It's been ages since I last referenced D3d errors but squares are either memory or bus related.

What u see is not uncommon.Gtav has similar artifacts.


Is the square always in the same place? What drivers btw?
What happens if u disable AA?

 

[ivicagmc]

sorry to hear , it'll get busy here soon. If those are Your +12V rails the PSU very well COULD be the problem...with age it gets worse.not to mention a 7970 eats Watts.

and no your shouldn't split the rails...I recommend NEVER to use those damn molex adapters, those are basically a way for the Card makers to get You to buy a new card sooner, by shitting your current one out.

OK, so to put graphics back on that rail and to lower powrlimit for 10%. You think that is enough?

 

[Pill Monster]

Btw don't flash a bios from another Tahiti onto yours. They aren't one size fits all. What vram is on the card?


Textibg from phobe so gghbb

 

[ivicagmc]

Not PSU. It's been ages since I last referenced D3d errors but squares are either memory or bus related.

What u see is not uncommon.Gtav has similar artifacts.


Is the square always in the same place? What drivers btw?
What happens if u disable AA?

NO it is not same place. And it is not always square. Sometimes are entire textures blinking.... Once it was texture over half of the screen.

 

[Ferrum Master]

Already did that and put some Prolimatec thermal paste

I did say not repaste... the bracket is usually bent round as a banana... the temperature diode doesn't show such errors.

If not... it is a bad RAM IC.

 

[human_error]

My MSI Gaming 290x is in for RMA at the moment because of the same exact issue (textures can be wonky, sometimes a square on screen flickering corruption, sometimes geometry corruption etc). I'd replaced my PSU and that had not fixed the issue for me. I have narrowed it down to one of the memory modules - if I increased the amount of VRAM usage it increased the likelihood of seeing an error to where I would certainly see one at very high memory usage (in low VRAM situations I would still see corruption, it would be less frequent though).

 

[ivicagmc]

Btw don't flash a bios from another Tahiti onto yours. They aren't one size fits all. What vram is on the card?


Textibg from phobe so gghbb

Hynix RAM. I did that bios flash because i found that was one of working solutions for artifacts and 7970 GHz edition and 280x are exactly same cards. Lataset 14.12 drivers...

 

[ivicagmc]

I did say not repaste... the bracket is usually bent round as a banana... the temperature diode doesn't show such errors.

If not... it is a bad RAM IC.

OK it is bad RAM ic. I can replace card but how to prove errors? IS memtestCL good for that? Right now I'm doing 2000MB 200 iterations to find errors. Unfortunately that is max ram for test.
And when you open card you always must re paste because of air pockets between chip and cooler

 

[Pill Monster]

Lol I just read the op agin. You're saying u get artifacts with overclocked vram and are worried the psu might be faulty after the system crashed at 1600mhz?

 

[ivicagmc]

Lol I just read the op agin. You're saying u get artifacts with overclocked vram and are worried the psu might be faulty after the system crashed at 1600mhz?

Not overclocked but underclocked. But in OCCT test i put ram on 1630MHz and lower powerlimit and there is no errors...

 

[Ferrum Master]

OK it is bad RAM ic. I can replace card but how to prove errors? IS memtestCL good for that? Right now I'm doing 2000MB 200 iterations to find errors. Unfortunately that is max ram for test.

I usually take screen shots and send them to choke on them with the card . There are not any good GPU RAM test programs anymore. They write in random addresses.

And for the 7970 and 280X, not all cards are the same there are XT2 and XTL versions of the chip... but in your case it doesn't change a lot... more loose memory timing could solve the problem, but that's a lot of hassle to handpick more crappy bios that carries slower timings...

BTW my original ram speed on 7970 is 1350MHz... and it is actually running on 1.5V only... Other cards like asus ran them on 1.55V by default... therefor they run on 1500MHz stock.

 

[Pill Monster]

It's not bad if the artifacts appesr with a 160

Hynix RAM. I did that bios flash because i found that was one of working solutions for artifacts and
7970 GHz edition and 280x are exactly same cards. Lataset 14.12 drivers...

The GPU's are the same. The bios configuration depends on more than that. You prob got a reference card. What's the batch number?


Tbh u don't nerdcet to worry toi much about vram overclocking, they alresdy have heaps of bandwith...

 

[ivicagmc]

It's not bad if the artifacts appesr with a 160The GPU's are the same. The bios configuration depends on more than that. You prob got a reference card. What's the batch number?


Tbh u don't nerdcet to worry toi much about vram overclocking, they alresdy have heaps of bandwith...

This is the card... http://www.gigabyte.com/products/product-page.aspx?pid=4845#ov
And it is same as http://www.gigabyte.com/products/product-page.aspx?pid=4290#ov
I found that bios is compatible and it is just incising GPU voltage for bost clock and thus incising stability. The say it worked for them but not for me. I got immediate artifacts in World of tanks. Usually it take 2-3 ours to get them. So I put latest bios F71 back.

 

[ivicagmc]

OK. memtestCL 0 errors after 200 iterations. I try that test on 1000 iterations later. I just need something to reproduce errors. Replacement is somewhat complicated for me this time.
And memtestCL 2000MB with unigine Valley working parallel with all graphics ram in business. Nothing, no artifacts, no errors.

 

[Bo$$]

sounding like power to me...

 

[ivicagmc]

sounding like power to me...

Should I make a pool out of this tread?

 

[MrGenius]

Oh, you mean a poll. Why bother? Probably do you more harm than good. A pool would be nicer. It's the PSU for the most part, if not entirely. Which couldn't handle this particular card even if it was new(and you could dedicate a full 2 x 18A +12V rails to it).

And, when you get a proper PSU(I recommend one with at least 60A +12V single rail if OCing), try upping the power limit to +20%. Then try running the Hynix as high as it will go. I run mine @ 1850MHz(fully stable no artifacts EVER) on my MSI 280X gaming. I've found that Hynix needs the OC to run well with the stock volts.

 

[ivicagmc]

One thing left. My wife PC before I find someone to borrow good PSU. I will try World of tanks there to see. It's Phenom 2 X3 lower TDP, less strain on PSU, will see...

 

[THE_EGG]

I had a rev2.0 Gigabyte 280X windforce just over a year ago. I experienced the same problems as you except for me it occurred in the first week or two of purchase. I sent it in for warranty and it was replaced with a new one. Gigabyte claim it was a bad batch of RAM on the card. I'd say send it for warranty and you should be able to get a replacement or a refund.

While you're at it I'd also change the PSU to a better designed one with one powerful 12V rail or something like that.

 

[ivicagmc]

There are artifacts on my wife PC. Than it is bad card.

 

[Pill Monster]

I've seen the same artifacts on a 7950 VaporX with high vram load, usually over about 2.5gb. I,m insure if this is related but remember reading tahiti has a max dedicated limit of around 2.2 the rest is reserved for buffers or something. In W10 i don't think this will apply because video memory will be virtualized. have to look at the GCN white papers to find exact numbers. I can't acess a pc atm only have my phone. Prob obvious by all the spelling errors lol