20 replies to this topic

[Dark Fact]

I think I've finally decided on the GPU I'm gonna buy--the ASUS GTX 670 DC2. But the first thing I'd like to do when I get it is check for defects/artifacts, and I believe stress testing is the way to do this.

Does anyone know the best tools for this task? (ie. Prime95 for CPU/RAM testing, you can leave it running and it will give you a report).

I've heard that EVGA's OC Scanner is popular, but does it work with non EVGA cards?
I've also heard that Furmark is not good because it pushes the card too hard, possibly leading to damage if left running for a while.

So, any recommendations, tips, advice?

[Vulpesveritas]

-->Furmark<--
http://www.ozone3d.net/benchmarks/fur/

[Dark Fact]

"I've also heard that Furmark is not good because it pushes the card too hard, possibly leading to damage if left running for a while."

I guess I could try it for a bit, carefully monitoring temps.. but I don't know, it seems kind of iffy!

[Shivus]

You just have to watch the temp until the card hits its highest point. My 6950 maxes out at 85-87 in furmark, while gaming it rarely breaks 76. Its maximum safe temp is 95. You don't want the card to be at 85-95 while gaming every day because that's how damage happens. But 85 during an hour or two stress test is perfectly safe, it's different for different cards of course.

Edited by Shivus, 24 June 2012 - 09:34 PM.

[Vulpesveritas]

Well, if you don't have good cooling, it can. It's meant to push the GPU to the true maximum it can be. Hence why.
You may want to manually set the fan to 100% before using it, but yes, you should keep an eye on it.

[silentD11]

EVGA and MSI afterburner work on cards they don't make. Shouldn't have a worrry, though I'd go wither after burner.

Furmark does NOT push a card to hard depending on what you are talking about. The VRMs (voltage regulation) on many cards will blow out before you fry the GPU, happens more on nvidia cards lately. Most programs only accurately get the GPU core temp and memory temp (don't worry about memory temp you have to try to fry the memory) but their VRM temps either don't exist or are wildly off.

You shouldn't be doing furmark with large OC's unless you are liquid cooling and it covers the VRMs properly (aka full cover block).

[Aznpersuasion89]

What about occt? I read the full stress test is insane.. http://www.ocbase.co...ex.php/software

Edited by Aznpersuasion89, 24 June 2012 - 10:30 PM.

[Dark Fact]

So I got the card and its cooler is wonderful. The fans on auto keep RPMs so low that you can't hear them. And this is while running the OCCT GPU stress test (in Error Checking mode, which supposedly generates a bit less heat). Temps are 71C with the case closed, 72C if I take the side door off.

I ran that for an hour and found no errors.

However, what I did notice is the occasional weird .. scan line (?) going up the screen. It is very faint, hardly noticeable. Running the EVGA OC Scanner stress tester also generates no errors, but it showed more of those scan lines. After enabling "heavy mode" under "power draw control" in OC Scanner, however, the faint flicker-like scan lines went away. It makes me wonder if there was just not enough power for the OC Scanner stress test under default settings?

Or is my GPU defective? D:

My power supply is a Seasonic X-560, considered to be a very high quality unit. I've seen posts by people indicating that a bad PSU can cause such scan lines, but I'm not sure if that was the cause here.

[Mu]

I could be wrong but that sounds like screen tearing to me. It's what happens when your framerate is going above the refresh rate your monitor can display. Nothing to worry about and if it bothers you ingame just turn on vsync. Or get a 120hz monitor, heh.

[killer panzer]

Dark Fact, on 25 June 2012 - 05:31 PM, said:

I ran that for an hour and found no errors.

However, what I did notice is the occasional weird .. scan line (?) going up the screen. It is very faint, hardly noticeable. Running the EVGA OC Scanner stress tester also generates no errors, but it showed more of those scan lines. After enabling "heavy mode" under "power draw control" in OC Scanner, however, the faint flicker-like scan lines went away. It makes me wonder if there was just not enough power for the OC Scanner stress test under default settings?

Or is my GPU defective? D:

My power supply is a Seasonic X-560, considered to be a very high quality unit. I've seen posts by people indicating that a bad PSU can cause such scan lines, but I'm not sure if that was the cause here.

Mu, on 25 June 2012 - 05:46 PM, said:

I could be wrong but that sounds like screen tearing to me. It's what happens when your framerate is going above the refresh rate your monitor can display. Nothing to worry about and if it bothers you ingame just turn on vsync. Or get a 120hz monitor, heh.

most likely it is a screen tearing, like mu suggests, it would be the simplest answer. while a bad PSU can cause these lines, i would expect that your unit needs more power. i propose that you replace the PSU not because it is "defective", but because you may be reaching the limits of your power usage. the rule of thumb i was taught was whatever your power needs, add 20 percent; if you have a computer that needs 1000 Watts, then get 1200. that solved my tearing problem when i added high volume fans, granted i added a 2nd PSU.

[Dark Fact]

Thanks for the responses! Unfortunately, screen tearing is not the issue as these wave-like patterns can be seen on the entire screen when the stress tester is in windowed mode or even minimized.

And I don't think the lack of power is an issue either because this strange glitch almost vanishes when I enable "heavy mode" in OC Scanner, which actually draws more power than when it is disabled. A high quality 560W PSU is also way more than enough to run a GTX 670.

Someone claims they fixed this issue by using another cable, and it does look like some sort of faint interference issue, but then why would it only happen under stress testing? That makes no sense! D:

EDIT: Actually, you may be right about it drawing too much power for the PSU... OC Scanner has a rendering delay option that reduces power draw. Increasing the delay slightly seems to reduce or fix the wave-flickering. But then why does increasing power draw also reduce (but not entirely fix) the flickering? @_@

I need to figure this out before the 15 day easy return/exchange period for my graphics card ends! I sure hope it's not the PSU, because RMAing that would be a pain in the butt.

Edited by Dark Fact, 25 June 2012 - 08:00 PM.

[silentD11]

If it's an actual card hardware problem odds are you will run into VRM and power issues far before you fry the GPU. Stock cards generally ship with crap tastic voltage regulation and craptastic voltage cooling. When you're paying $$$$ for the MSI lightning edition or things like that you're paying the extra $$$$ for the better voltage regulation.

GPU's generally aren't as temp sensitive as CPU's, but the VRM's themselves are constantly causing issues.

When you stress test it those VRMs are going to heat up and fail far before the GPU itself. And worst off, most programs do not measure the GPU VRM temp, just the core and maybe the memory, so short of actual thermal probes you don't know what's going on there.

[cipher]

Dark Fact, on 24 June 2012 - 09:27 PM, said:

"I've also heard that Furmark is not good because it pushes the card too hard, possibly leading to damage if left running for a while."

You can't push a card "too hard". A video card controls its own hardware, so benching software can't really push a card any harder. The only exception would be for software that can change the voltage or clocks of the hardware, but FurMark doesn't do that.

The biggest reason people think this is because they're running too much voltage, pushing their card too much for overclocks, or using inadequate cooling when they attempt to stress test their card(s).

As Vulpes linked, FurMark is a great benchmarking and stress testing tool. I use it all the time on new installs or video cards or just to troubleshoot.

[silentD11]

cipher, on 26 June 2012 - 07:16 AM, said:

You can't push a card "too hard". A video card controls its own hardware, so benching software can't really push a card any harder. The only exception would be for software that can change the voltage or clocks of the hardware, but FurMark doesn't do that.

The biggest reason people think this is because they're running too much voltage, pushing their card too much for overclocks, or using inadequate cooling when they attempt to stress test their card(s).

As Vulpes linked, FurMark is a great benchmarking and stress testing tool. I use it all the time on new installs or video cards or just to troubleshoot.

Furmark will load up a card completely, that's the point of it. However a lot of cards are not built to be run fullbore. There was a long period of time where furmark was killing cards left and right because of failing VRMs.

[cipher]

silentD11, on 26 June 2012 - 07:32 AM, said:

Furmark will load up a card completely, that's the point of it. However a lot of cards are not built to be run fullbore.

Yup, it loads it completely. But what cards are "not built to be run" at 100% GPU usage? Or do you mean non-gaming video cards, like the ones with passive coolers from the value line AMD or NVIDIA GPUs?

[silentD11]

cipher, on 26 June 2012 - 08:27 AM, said:

Yup, it loads it completely. But what cards are "not built to be run" at 100% GPU usage? Or do you mean non-gaming video cards, like the ones with passive coolers from the value line AMD or NVIDIA GPUs?

Even some gaming cards blow their VRMs. This was a big problem in the nvidia 8800 series with some cards. The stock configurations cheap out on the cooling, so in some cases there was nothing on the VRMs. That's what spurred companies like EVGA to add extra cooling on some versions, and MSI and others to create versions with beefier VRMs and cooling.

It's generally that reference cards have shoddy vrms and cooling. The GPU and memory can run full bore all day, your VRMs will crap out. Gaming vs htpc vs workstation card doesn't matter, it's companies cheaping out on the design. Because outside of people folding 24/7, nobody runs their cards full load like that.

[cipher]

silentD11, on 26 June 2012 - 09:04 AM, said:

It's generally that reference cards have shoddy vrms and cooling.

Sorry, have to call BS on that generalization. I'm sure there are some cards out there with cruddy VRM cooling, etc., but not all reference cards are like that. Here's a disassembled reference GTX 680 as proof...

Edited by cipher, 26 June 2012 - 09:20 AM.

[silentD11]

Now compare that cards VRMs to the MSI Lightnings and you'll see what I mean.

Games generally don't run cards 100% load. The entire point of the super OC cards is better VRMs to gang busters them.

[cipher]

silentD11, on 26 June 2012 - 09:51 AM, said:

Now compare that cards VRMs to the MSI Lightnings and you'll see what I mean.

Games generally don't run cards 100% load. The entire point of the super OC cards is better VRMs to gang busters them.

Do you know what's cooled and not cooled in that picture? All of the mosfets are cooled.

VRMs are comprised of inductors (ie: chokes), mosfets (or another type of transistors), and capacitors. Obviously the capacitors don't need cooling. The mosfets absolutely need cooling, so they make thermal contact with the huge heat spreader. The inductors (the large gray squares) don't always need cooling, depending upon their type and composition. Many video cards these days use low-temp inductors. Even then many inductors are rated up to 200C max operating temps, so that's not typically an issue. Mosfets usually blow before anything happens to an inductor (which is just a coil).

And yes, games can run cards up to 100%, depending upon the card, the game, and the settings. But they don't always run them at 100% all the time. That's the difference.

Edited by cipher, 26 June 2012 - 10:01 AM.

[silentD11]

Not all reference designs have good VRM cooling, not all reference designs have beefy VRMs. This is a hit or miss issue, but you will still most likely fry the VRM before you burn out the GPU.

The MSI lightning, ASUS matrix/mars, gigabyte super OC all ship with a completely redone VRM and better cooling for it, that's why they cost more.

As someone that LCs all my cards, I don't really have problems with it. But when benching the cards on xtremesystems and other places I've sure as **** burned through the VRMs of multiple reference cards.

And yes I know what makes up a VRM, I'm a bencher.