Factory OC'ed HD4890 artifacts unless I underclock it to reference

[RevOne]

Greeting TH Forum

Here's the card: XFX HD 4890 Extreme Edition
http://www.hardocp.com/article/2009/10/05/xfx_radeon_hd_4890_extreme_review/

So my GPU is not the latest and greatest by any stretch of the imagination, but it's all I've got atm 😀 Anyway, here's the skinny:

Details:
- Bought it secondhand off ebay back in (iirc) June 2011.
- Original owner did not register for the XFX double-lifetime warranty, so it couldn't be transferred over when I made the purchase.

Problem:
- Everything was working perfectly with no issue until I made the update to the Catalyst 12.2 drivers a couple months ago. Shortly after installing those drivers, I was getting heavy artifacting, freeze ups, and BSOD's (with an "atikmpag.sys" error). Even at idle, the screen would slowly start to display yellow pixels more and more until it would inevitably and eventually crash or lockup. Youtube videos and gaming was like asking for an instant BSOD. Rolled back to earlier drivers and issue persisted.

Troubleshooting:
- Took it apart. Inspected & cleaned the fan/cooler unit and PCB carefully. No burnt, swollen, or otherwise broken parts. Removed old thermal paste and applied fresh Arctic Silver 5 to the memory chips and reassembled.
- Opted to do a completely fresh install of Windows 7 x64 thinking maybe there was a corrupted driver installation that Driver Sweeper/Driver Cleaner Pro and software like that couldn't fix.
- Fresh install of the OS and a roll back to Catalyst 11.12 drivers... and everything seemed perfectly fine again. Ran cooler because of the new (better) thermal paste. Even ran and passed Furmark testing with no errors. For a while, at least.
Then the artifacting started happening again about a week or so later.
- Finally, I go into AMD Overdrive and underclock the card about 25MHz on both the core and memory.... and voila!... Haven't had issues running the card at all since (Now running the latest Catalyst 12.4 drivers). Can play all my games (BF3, Deus Ex: HR, Batman: Arkham City), watch Youtube videos... everything, just fine. But if I bump it back up only 25MHz on each slider to it's default (factory OC'ed) frequencies, the artifacting/BSOD's/lock-ups return.

This is a factory overclocked card @ 875 Core/ 975 Memory. However, if I underclock it to reference HD 4890 frequencies (850 Core/ 950 Mem), it works perfectly fine. I can't speak for the previous owner, but I never overclocked the card beyond what the manufacturer already did (left it at stock). And it worked great until I installed those drivers. Card temps never peaked above 75° C at full load.

I wouldn't think a measly 25MHz could yield such a major difference in stability. Does anyone have any words of advice to offer?
For instance:
1) Can a card's bios become corrupted and cause such issues?
2) Can bad drivers corrupt a card's vram?
3) Can overclocking other parts of your system adversely affect your GPU (HT Link; RAM timings, etc.)?

Any help would be greatly appreciated Thanks in advance!

 

[wavetrex]

Most likely the previous owner run it into a poorly ventilated case, and it run constantly at high temperature.

48xx series are notoriously hot ( I know, I still use one )

Try setting the fan on manual setting ( around 45% should do it ), it will be more noisy ( all the time ) but it will cut 20-30 degrees of the temperature, probably allowing you to run at default settings.

Also, use CCC to set up a "low clock" and "high clock" settings, with keyboard shortcuts, so you underclock (and lower fan speed) when you're in windows desktop and overclock (with high fan speed) when gaming. It's just a press of a key before starting a game, and another one after exiting it, not much work to do.

My 4870 "defaults" at 80deg, but with fan settings it runs well below 60deg, and I can overclock it 5% more without crashes. I can even put extra 9% (to stock) speed but only with fan at 80%, however it's noisy as hell. 45% works nice, not too loud.

 

[RevOne]

Thanks wavetrex

Indeed... these 48xx series cards could definitely cook you a nice golden brown grilled cheese sandwich without a problem 😀

When I first got the card and began testing, I would leave the fan on Auto and as you mentioned with your 4870, it would hit 80°C - 90° C (mostly during Furmark burn-in/benchmarking, regular gaming temps were lower) at default Auto fan settings under full load. Ever since, I've been running it with two separate hot-key profiles - one with the fan at 50% (which isn't too bad... my speakers are enough to drown it out) and the other at 70% for when I'm gaming (which is annoyingly loud). The card generally idles at around 50°C. Ambient room temps are a consistent 65 - 72 (at most) degrees F. Case is sitting in a well-ventilated area as well.

The card was also in pristine condition and ran perfectly when I received it from the original owner and only started showing troubles when I installed and around the time those Catalyst 12.2 drivers were released (I want to say maybe three? months ago). I'm using the following power supply to run it along with a single 1TB HDD, a 95w Phenom II X4, one optical drive, and two system fans: http://www.newegg.com/Product/Product.aspx?Item=N82E16817371015 . I'm fairly certain it's been getting adequate and clean power. Not really sure what's going on with it... :/

It's not such a huge deal for me to personally run it at reference 4890 specs (honestly, that "whopping" extra 25 MHz on the core & mem are not being missed), but it'd be nice if I could get it working per manufacturer's specifications at least. I've even gone as far as contemplating reflashing the bios to see if that fixes matters. I found the file on TPU... but I'm not hopeful that will yield any positive results.

 

[RevOne]

Well... I tried reflashing the card with the stock bios for this model provided by TPU here: http://www.techpowerup.com/vgabios/index.php?page=1&architecture=ATI&manufacturer=XFX&model=HD+4890&interface=PCI-E&memSize=1024

... and same results. :/

So, since I never really intended to overclock this card anyway, I used techpowerup's ATI Winflash and RBE utilities to extract the bios and changed the highest default clocks to 825MHz Core/ 925 MHz Memory (a total 50MHz Core/ 200MHz Memory decrease, just to be on the safe side). Performance and stability are rock solid at these frequencies; as I've run it through a gamut of testing with no issues at all and have been running it at these reference card frequencies for months now with no issues since the problem first began. So while it will never be an "overclocker-friendly" part and the card physically *looks* like an XFX 4890 "Extreme Edition" GPU, it now "acts" and is essentially exactly like a standard, reference 4890 (which honestly, if it gave me 1 to 2 more FPS before at factory OC'ed settings, I'd be surprised). Besides, if you read that review I linked to in the first post, hardocp mentions that these XFX "Extreme" cards are pretty much reference 4890's anyway, only with a very mild overclock... so it's not surprising.

Only other thing I did go ahead and modify in the bios since I was at it was the fan scaling profile settings ; which were copied from the top-tier XFX "XXX" version of the card's bios (so it keeps the card cooler under Auto fan settings in CCC). Voltages were exactly the same between the two parts, so it seemed like a safe enough bet. Now with the Auto fan option in CCC, it's getting 59° C idle & 77°C load. I'd say a good average 7° drop in temps now compared to using the stock bios' Auto fan profile.

So all in all - still isn't a bad outcome at all considering it's only a mild underclock and I'm not an overclocker anyway. Better than needing to totally replace the card prematurely. And it's working beautifully . If all you need is a good, solid performing "almost" reference 4890 card and have no intentions of overclocking, this card will do it all day; no questions asked. Will still easily beat out any HD 4870 (w/ 725 Core/ 900 Memory) and even the HD 5770/6770 by a significant enough margin at pretty much everything... even with the underclock. So... really can't complain at all.

Thanks for the assist, too, by the way Wavetrex