Artifacts in various games? Tested across multiple drivers. Video inside. (Update)

[Dankk]

Update 2: I didn't get around to testing this issue on 12.4, until just today. Apparently I was indeed correct that the artifacts didn't start showing up until 12.6. After reverting back to 12.4, I'm having an extremely difficult time replicating these artifacts, in fact I have not deliberately been able to replicate any at all (besides one random one in DNF, but that's in a 30-minute span of playtime). This leads me to believe that this actually might be a driver problem.

---

When I got my Sapphire Radeon HD 7870 back in April, I don't remember having artifacts like this. I think it's been Catalyst 12.6 onward when I begun to notice artifacts in various games. (That may have been when the artifacts actually started, OR it's entirely possible they were there all along and I just happened to not notice for a while.)

They all look the same: Brief, intermittent flashes of black polygons smeared across the screen. I've never seen them in different colors before... only black. I've tried multiple drivers; 12.6, 12.7 beta, and 12.8, but no luck. I was crossing my fingers that it was just a driver problem and 12.8 would fix it, but... no dice.

Here's a video I recorded with FRAPS. I'm using Duke Nukem Forever as an example because the artifacts are most obvious here, and I can easily reproduce them. The footage is slowed down to 50% speed so it is more noticeable. (The artifacts are noticeable at 6 seconds and 11 seconds in.)

http://www.youtube.com/watch?v=bIPLrkozKoA

Update: Recorded another video. Artifacts are much more prevalent in this one. Also, looks like I'm probably wrong about them only being black; I think I'm seeing different colors here:

http://www.youtube.com/watch?v=864iWytOZzU

There's several other games I'm getting the same artifacts in; they're extremely subtle, not as noticeable as in Duke Nukem Forever, but they're there. Ridge Racer Unbounded, Payday The Heist, Dead Island, and Sleeping Dogs have all exhibited these brief black artifacts to some extent.

Edit: I should mention that I do completely clean installs of my drivers. I use Driver Fusion in safe mode and reboot after every step when updating. Also, I have not overclocked my card at all. Temperatures are perfectly fine... 35 idle, maybe 70 when playing a game.

Any ideas? Is my card dying?

 

[SlowSpyder]

There have been times that I have seen things like that when I overclocked too high. Are you overclocked at all?

 

[Dankk]

There have been times that I have seen things like that when I overclocked too high. Are you overclocked at all?

No, not overclocked at all.

Sorry, that's another important piece of information I should probably put in the OP.

Should I try underclocking?

 

[SlowSpyder]

I would, yes. If the problem clears up, then I would think the GPU/card is deffective. Try underclocking the GPU, try underclocking the memory. If nothing else, it may help rule that out.

 

[Dankk]

I would, yes. If the problem clears up, then I would think the GPU/card is deffective. Try underclocking the GPU, try underclocking the memory. If nothing else, it may help rule that out.

Unfortunately I'm not familiar with overclocking (or underclocking, for that matter). I went into Catalyst Control Center, open up Overdrive, and lowered the the Core and Memory clocks from 1000 and 1200 all the way down to 400 and 600 (respectively). Went back into Duke Nukem Forever and still got artifacts though.

Is there something else I'm supposed to do? I've heard of "under-volting" video cards, is that different? Should I try that?

 

[chimaxi83]

Unfortunately I'm not familiar with overclocking (or underclocking, for that matter). I went into Catalyst Control Center, open up Overdrive, and lowered the the Core and Memory clocks from 1000 and 1200 all the way down to 400 and 600 (respectively). Went back into Duke Nukem Forever and still got artifacts though.

Is there something else I'm supposed to do? I've heard of "under-volting" video cards, is that different? Should I try that?

Under volting really shouldn't make a difference if you already down clocked and you're still artifacting. Sounds like you've got a handle on drivers and clean installs, so your card is probably bad.

 

[CarpeDiem99]

Looks like a faulty card, RMA it

 

[Dankk]

Just an update: I have a friend who suggested that I re-seat my video card. He also suggested that I "power-cycle" my computer, on the off-chance that static buildup is somehow causing corruption.

It didn't work. For the record, this is what I tried, in order:

1) Completely unplugged the computer, flipped the switch on the back of the PSU.

2) Layed down the PC on it's side, unplugged and removed the graphics card. Set it aside.

3) Gently blow some compressed air inside my PC to dust it out (as I usually do whenever I open it up)

4) Held down the power button for ~30 seconds.

5) Firmly re-seated the video card, tightened the screws and plugged in the power cables.

6) Held down the power button for another ~30 seconds (this time with the graphics card plugged in).

7) Let the computer sit there for about an hour before plugging it back in and turning on.

Again, nothing. Booted up and ran Duke Nukem Forever with my original clock settings. Still getting intermittent artifacts.

Looks like a faulty card, RMA it

How is Sapphire's RMA service? I got the card from Amazon, and the store return-window expired a long time ago, so I'd have to go through the manufacturer.

My last RMA experience was horrible. I had an XFX 6950 that had a really bizarre problem: it would display multi-colored artifacts, but only in DX11 games with tesselation. I returned the card to XFX twice thoroughly explaining the problem, and they messaged me back multiple times saying "we've tested the card and have found no problems, so something must be wrong with your computer." I tested the card - probably more extensively than they did - and 100% objectively deduced that it was the card's fault, and I even sent them screenshots. They eventually sent me a replacement card but still refused to recognize the problem... they only said "we've tested the card," with no further information on what they actually tested it with. Suffice to say, I'm not buying anymore XFX cards.

I don't feel like sending my card to Sapphire if they're going to reply back with a generic "We've tested your card and found no problems" message.

Any other ideas? Unless someone can think of another possible fix for the artifacts, I'll RMA it. If nothing comes up by next Monday then I'll return it.

 

[CarpeDiem99]

I googled Sapphire rma experience, seems to have mixed results, with Canadian users having to pay a fee.

I didn't see any complaints about getting the same card back though, so you shouldn't have that problem, and being in the US, you probably don't have to pay a processing fee either.

 

[RussianSensation]

Before you RMA the card, increase GPU voltage in MSI afterburner. Give it 0.05-0.75V on top of stock. If that doesn't fix the issue, and you try older drivers such as 12.4-12.5, then it's probably a faulty card. Sometimes a poor driver installation can corrupt drivers in the registry. It's rare but it happens. Maybe you can try the card at a friend's computer. If the same artifacts occur, I'd RMA.

Also, before that use MSI Kombuster in MSI Afterburner and run the GPU Burnin test for 10-15 minutes. It should completely freeze if the card is unstable.

Take a look at this video and it should guide you through what you need to do:
http://www.youtube.com/watch?v=85sox46roxM

 

[Dankk]

Before you RMA the card, increase GPU voltage in MSI afterburner. Give it 0.05-0.75V on top of stock.

Installed MSI Afterburner. Starting at 1118mV, I increased the voltage three or four different times, loading up Duke Nukem Forever each time to see the results. It didn't help.

If anything, increasing the voltage seemed to make the problem worse, not better. I went as far as 1156mV and tested it out, and it looked like the artifacts were almost certainly more prominent than before. Decided not to push it any further than 1156mV.

Also, before that use MSI Kombuster in MSI Afterburner and run the GPU Burnin test for 10-15 minutes. It should completely freeze if the card is unstable.

I ran Kombuster GPU burn-in for ~15 minutes in Direct3D 9 mode, no crashes. Ran it again for another ~15 minutes in OpenGL 2 mode; again, no crashes. On stock settings it ran perfectly smoothly.

If that doesn't fix the issue, and you try older drivers such as 12.4-12.5, then it's probably a faulty card. Sometimes a poor driver installation can corrupt drivers in the registry. It's rare but it happens.

I'm going to bed now, so I'll try this tomorrow. I'll uninstall 12.8 and revert all the way back to 12.4 to see what happens. I'll report back if it makes a difference or not.

In the meantime, I created a support ticket with Sapphire. Didn't mention anything about an RMA yet; just clearly and concisely explained the issue to them along with a screenshot. We'll see what they have to say.

I'm worried that this problem is so subtle that they won't do anything about it. Again, I'm only using Duke Nukem as an example because the artifacts are so obvious in that game; but I know they're appearing in other games too. Sleeping Dogs, for example, appears flawless 99.9% of the time, but there are those few occasions when I know I saw a brief dark flash in the corner of the screen. A rogue artifact that only lasts for a fraction of a second before disappearing again.

To the trained eye it's noticeable, but the question is if Sapphire will pay that same amount of attention that I do.

 

[RussianSensation]

Download HWInfo64 and click Sensors, and take a look at your GPU VRM temperatures, GPU temperatures, +3.3V and +12V PSU load ratings under GPU load. What you are looking for here is the Minimums for the voltages and Maximums for the temperatures. If 3.3 and 12V rails dip way under spec, the PSU may not be providing stable power. This is prob not the case but just want to rule out other factors.

Also, this is where we get to see if your card's GPU voltage is working properly as per your MSI Afterburner adjustments. You set it at 1.156V but does HWInfo64 report it as 1.156V or something ridiculous like 1.08v? Did you check that the PCIexpress connector from the PSU is properly flush with the card? You should have a click when you plug those into the videocard.

 

[Dankk]

Opened up HWiNFO64 and looked at +3.3v and +12v values, both when idling and when playing DNF. First of all, temperatures look fine. As for PSU load ratings,

My +12v normally sits at 12.288v, however it dipped down to 12.192v when playing Duke Nukem Forever.

My +3.3v normally sits at 3.408v, however it dipped down to 3.376v when in-game.

Is that bad?

Oh, and I recorded another video. I seem to have found the sweet-spot for artifacts. When I stand in a particular spot, it goes nuts. *sigh*

http://www.youtube.com/watch?v=864iWytOZzU

I also think I might've been wrong about them only being black-colored. I'm definitely seeing different colored corruption in there.

 

[PrincessFrosty]

What are the temps exactly? These kind of artefacts are typical of the GPU not calculating things correctly almost always due to overheating.

Make sure to graph temps under extended play time and post the peak values.

 

[Dankk]

What are the temps exactly? These kind of artefacts are typical of the GPU not calculating things correctly almost always due to overheating.

Make sure to graph temps under extended play time and post the peak values.

When idling, GPU temp sits around a cool 30 or 35 C (depending on the time of day and if I have my AC on or not).

When running MSI Kombustor burn-in test for 15 minutes, it maxed out at 70, tops. No higher than that. In regular games it's a bit lower... in Duke Nukem Forever, for example, it maxes out at about 62 or so. Not hot at all.

Unless the temperature sensor is totally broken, I'm confident my temps are fine.

RussianSensation mentioned both "GPU temperature" and "GPU VRM temperature" though... I take it there's more than one temperature I should be looking at? I'm at work right now so I'll have to wait until I get back home to double-check that.

 

[RussianSensation]

Your +3.3 / 12V look on the high side but it's not a major problem. I think your card may just be bad. Can you try it on your friend's system to confirm before RMA?

 

[Dankk]

Your +3.3 / 12V look on the high side but it's not a major problem. I think your card may just be bad. Can you try it on your friend's system to confirm before RMA?

Thank you for the clarification!

HWiNFO64 didn't appear to have a reading for VRM temperature. Booted up the latest version of GPU-Z though, and it reports two different VRM temperatures. So I used that instead.

While idle, both values show 34 C. Not bad.

However, when I ran DNF for a few minutes, VRM temperatures 1 and 2 maxed out at 89 and 127 C, respectively. Is that bad?

Looking at line graphs, it looks like the temperature spiked upward in very short increments to reach those values. Maybe that's somehow correlated to when the artifacts appear? (Or am I just making stuff up?)

 

[Dankk]

I should add: That last GPU-Z screenshot was taken after I'd only set it to record max values, and then run the game for a few minutes before exiting back out again. Because I was running the game full-screen, I didn't see a real-time side-by-side comparison... only a before-and-after.

I decided to run the game again in windowed-mode with GPU-Z running directly alongside it. Turns out, on average, the VRM temperatures only sit at around ~47 C most of the time. However, it intermittently spikes up to 60, 80, or even the 120 range.

Timing-wise, I didn't see a direct correlation between the temperature spikes and artifacts, but something definitely seems up with those VRM temps. Someone tell me if those are normal or not.

 

[chimaxi83]

Wow, 127 C. In my experience, GPU-Z temperature readings tend to be off, along with current and voltage readings. I rely on HWiNFO64, HWMonitor, and Afterburner. Might be a good idea to check that the heatsink assembly is properly installed and tight.

 

[RussianSensation]

Something is definitely going on with those VRM temperatures. The threshold I belive is only 120*C or 125*C. You are getting into danger territory. Does your card have a metal heatsink on top of the VRM circuitry? Check if it's sitting flush with the card. Take off a case fan and put it to blow on the side of the GPU. See if those VRM temperatures fall below 120*C.

 

[Dankk]

Wow, 127 C. In my experience, GPU-Z temperature readings tend to be off, along with current and voltage readings. I rely on HWiNFO64, HWMonitor, and Afterburner. Might be a good idea to check that the heatsink assembly is properly installed and tight.

Unfortunately, GPU-Z is the only hardware monitoring program that shows me VRM temperature. HWiNFO64, HWMonitor, and SIW don't appear to have entries for VRM (at least, not that I can see). I don't see anywhere in Afterburner that gives me VRM temperature readings either.

Something is definitely going on with those VRM temperatures. The threshold I belive is only 120*C or 125*C. You are getting into danger territory. Does your card have a metal heatsink on top of the VRM circuitry? Check if it's sitting flush with the card. Take off a case fan and put it to blow on the side of the GPU. See if those VRM temperatures fall below 120*C.

Opening up video cards is a bit beyond my comfort zone. However, I have a friend who's a lot more technically-inclined than I am, and he's familiar with disassembling them. Has the right tools for it too. I might take it to him.

This might be pushing it, but: Would re-installing DirectX solve the problem?

 

[Dankk]

Here's another GPU-Z screenshot I took after playing Sleeping Dogs for about an hour. Yeah... that definitely doesn't look right.

 

[RussianSensation]

I am not an expert but those temperatures don't make any sense. Something is wrong with the videocard VRM circuitry/sensor.

Can you please link the exact card you purchased? This?

http://www.newegg.com/Product/Produc...82E16814102983

You should send both of those GPU-Z pictures to SapphireTech and ask them if you should issue an RMA. That looks like you have something faulty with the card's VRM sensor or the VRM's are overheating like crazy.

For example TYCO electronics VRM temperatures are rated with max. operating temperature of 125°C — VRM 10.x Series

The 125*C number tends to be the upper end of the safe range from what I read on AMD cards. But that's from anecdotal evidence, not engineering boundary. Just giving you an idea.
http://www.hardwarecanucks.com/forum/video-cards/51688-what-safe-vrm-temperatures.html

It's difficult to find accurate data since it's hard to know the manufacturer of those specific VRMs. Either way 255*C is so far past the reasonable limit, you know you have a hardware/sensor problem. The fact that the card is artifacting and VRMs are running super hot sounds like they are overheating to me. There is no way you can solve that with a simple 120mm fan. I'd RMA that card.

 

[Dankk]

I am not an expert but those temperatures don't make any sense. Something is wrong with the videocard VRM circuitry/sensor.

Can you please link the exact card you purchased? This?

http://www.newegg.com/Product/Produc...82E16814102983

You should send both of those GPU-Z pictures to SapphireTech and ask them if you should issue an RMA. That looks like you have something faulty with the card's VRM sensor or the VRM's are overheating like crazy.

For example TYCO electronics VRM temperatures are rated with max. operating temperature of 125°C — VRM 10.x Series

The 125*C number tends to be the upper end of the safe range from what I read on AMD cards. But that's from anecdotal evidence, not engineering boundary. Just giving you an idea.
http://www.hardwarecanucks.com/forum/video-cards/51688-what-safe-vrm-temperatures.html

It's difficult to find accurate data since it's hard to know the manufacturer of those specific VRMs. Either way 255*C is so far past the reasonable limit, you know you have a hardware/sensor problem. The fact that the card is artifacting and VRMs are running super hot sounds like they are overheating to me. There is no way you can solve that with a simple 120mm fan. I'd RMA that card.

Thanks for the response. Sapphire already responded to my support ticket, asking me to forward them a copy of the purchase receipt so they can issue an RMA. In the two messages I sent them, I attached a DNF screenshot, and I also attached a GPU-Z screenshot (the one displaying 255 C for both values).

As for video card model: You're close. It's the non-OC version of that. SAPPHIRE Radeon HD 7870 GHz Edition 2GB 256-bit GDDR5 PCI Express 3.0 x16 HDCP Ready CrossFireX Support Video Card (11199-00-20G)

Interesting observation: 255 is actually the maximum value representable with 8 bits. That may have been as high as GPU-Z was allowing it to go.

But yeah, I'm definitely thinking RMA. Good thing I have another video card I can use in the meantime.

 

[RussianSensation]

I hope they sort this out for you and you have a smooth RMA. Take a picture/write down the serial # of your card to make sure they don't send you the same card back!