What is the best way to check for stability on a GPU OC?

vollman1

Member
Mar 8, 2012
62
0
0
Title says it all:

What is the best way to check for stability on a GPU OC?

What do you run and for how long? Or do you just play and if there is a crash, lower your settings?
 

Jhatfie

Senior member
Jan 20, 2004
749
2
81
Honestly gpu intensive games like Crysis2 at heavy detail settings seem to be the most reliable stress test for me. I've been able to loop benchmarks (Heaven, 3dMark11) and gpu burning utilities for hours and have no issues, then play 15 minutes of Crysis at the same settings and crash or see artifacts because of the OC being unstable.
 

Mem

Lifer
Apr 23, 2000
21,476
13
81
Title says it all:

What is the best way to check for stability on a GPU OC?

What do you run and for how long? Or do you just play and if there is a crash, lower your settings?

Try FurMark http://www.ozone3d.net/benchmarks/fur/ ,thats what I use.


FurMark is a very intensive OpenGL benchmark that uses fur rendering algorithms to measure the performance of the graphics card. Fur rendering is especially adapted to overheat the GPU and that's why FurMark is also a perfect stability and stress test tool (also called GPU burner) for the graphics card.
 

MrK6

Diamond Member
Aug 9, 2004
4,458
4
81
Honestly gpu intensive games like Crysis2 at heavy detail settings seem to be the most reliable stress test for me. I've been able to loop benchmarks (Heaven, 3dMark11) and gpu burning utilities for hours and have no issues, then play 15 minutes of Crysis at the same settings and crash or see artifacts because of the OC being unstable.
Bingo. Most benchmarks or "stress tools" like OCCT or Furmark don't completely load all the GPU's components. Programs like Furmark are just power viruses that do little more than test the capacity of your GPU heatsink. Both AMD and nvidia now include throttling measures in their drivers to even prevent these power virus programs from maxing out the cards. Therefore, playing a modern game is going to be your best test for stability.
 

vollman1

Member
Mar 8, 2012
62
0
0
I have heard people saying that programs like Furmark are 'card killers.' Never really knew what they were referring to. Maybe causing the card to overheat, is that what you mean?
 

chimaxi83

Diamond Member
May 18, 2003
5,457
63
101
Those programs basically make your card draw maximum power and puts 100% load on your VRM's, which will cook and probably cause reflow of solder if they're not properly cooled.
 

MrK6

Diamond Member
Aug 9, 2004
4,458
4
81
I have heard people saying that programs like Furmark are 'card killers.' Never really knew what they were referring to. Maybe causing the card to overheat, is that what you mean?
In essence, yes. Before OCP (over-current protection) was heavily implemented into both hardware and software, power virus programs like these could cause the card to draw so much current as to kill components/circuitry in the card. Now it's near impossible to do so.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,570
10,202
126
In essence, yes. Before OCP (over-current protection) was heavily implemented into both hardware and software, power virus programs like these could cause the card to draw so much current as to kill components/circuitry in the card. Now it's near impossible to do so.

Only because GPU power circuits are under-engineered. Imagine if CPUs or motherboards burned out under "full CPU load", because they were designed to imagine that nobody would run their PC at full load.
 

amenx

Diamond Member
Dec 17, 2004
4,299
2,628
136
I hope people do away with the belief or view that GPU's need to be stress tested in the same manner as CPUs. I would think nothing of using stessful CPU programs like Intel burn test or OCCT for CPUs. But when it comes to GPUs, I would rather find that an OC is unstable thru actual game crashes rather than these killer stress programs like Furmark or OCCT which no game in existence would come close to stressing. Even more silly is that I've actually seen a couple of folks in other forums try to use these programs for hours on a GPU like they assumed is done on a CPU.

For a GPU, Crysis is a good stability test for me. As well as a few runs of 3dmark. Sometimes a game can be unstable due to the CPU OC, even if the CPU passes stress tests. Happened to me in a couple recent games (Deus EX HR and Skyrim).
 
Last edited:

skipsneeky2

Diamond Member
May 21, 2011
5,035
1
71
I have found myself heavy game loading is the best way to bring out a failure in a oc...either its piss poor performance instability a hard lock or just artifacting.

Me my testing grounds is maxing out a game like BF3 for example with all aa options,this way the memory on the gpu gets used to its full potential,max the settings and run around in the game,it may be slow but any instability issues should arise quickly.

I used to use programs like occt and furmark,nothing but a waste of time,i could pass either for hours on end,but BF3 or a game always shows instability within 20 mins guaranteed.
 

fstime

Diamond Member
Jan 18, 2004
4,382
5
81
How come my 100% stock Radeon 7950 crashes and makes my screen turn black after running OCCT or Furmark after 3-5 minutes yet will play a game o/c'd or loop Heaven while o/c'd for as long as I want with no artifacts?

Is there a safety feature shutting the card down? Someone said above these aren't true tests, just power hungry viruses. But why make them if they cant even run on a stock card?
 

chimaxi83

Diamond Member
May 18, 2003
5,457
63
101
How come my 100% stock Radeon 7950 crashes and makes my screen turn black after running OCCT or Furmark after 3-5 minutes yet will play a game o/c'd or loop Heaven while o/c'd for as long as I want with no artifacts?

Is there a safety feature shutting the card down? Someone said above these aren't true tests, just power hungry viruses. But why make them if they cant even run on a stock card?

Sounds like a heat issue which normal gaming or Heaven won't reproduce.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,570
10,202
126
How come my 100% stock Radeon 7950 crashes and makes my screen turn black after running OCCT or Furmark after 3-5 minutes yet will play a game o/c'd or loop Heaven while o/c'd for as long as I want with no artifacts?

Is there a safety feature shutting the card down? Someone said above these aren't true tests, just power hungry viruses. But why make them if they cant even run on a stock card?

Probably your power-supply can't handle it.

Modern GPUs do have some power-usage-containment features, but generally, you don't end up with a black screen, you just end up with slower performance.

But if the PSU shuts down, you will notice it.
 

MrK6

Diamond Member
Aug 9, 2004
4,458
4
81
Those Furmark type programs won't damage recently manufactured cards then?
No, they won't, but that also eliminates their "effectiveness" (which is a ruse in itself). Case in point, don't use Furmark or similar programs.
Only because GPU power circuits are under-engineered. Imagine if CPUs or motherboards burned out under "full CPU load", because they were designed to imagine that nobody would run their PC at full load.
Not really true. I believe that by design, GPU's are easier to exploit via a power virus, so their loads are multitudes higher than would be on a CPU. Furthermore, a video card's PCB is much smaller than that of a motherboard, so they make do with what space they have.
How come my 100% stock Radeon 7950 crashes and makes my screen turn black after running OCCT or Furmark after 3-5 minutes yet will play a game o/c'd or loop Heaven while o/c'd for as long as I want with no artifacts?

Is there a safety feature shutting the card down? Someone said above these aren't true tests, just power hungry viruses. But why make them if they cant even run on a stock card?
Some 7950 models have the bare minimum cooling on the VRM's. While it's more than adequate for stock settings, a power virus will exploit this deficiency and crash the card due to the high temps. Don't run Furmark or similar programs anymore as you will eventually damage your hardware.
 

fstime

Diamond Member
Jan 18, 2004
4,382
5
81
I have the Asus 7950 DC II which is a 3 slot cooler. My PSU is a Corsair TX650 which has Seasonic internals and should be more than enough for a single cars setup.

Does this sound like a faulty card or poorly designed card in general? Really considering returing it. Like I said, I can play Skyrim for hours and my temps wont breach 65C with an idle of 30C and this is while overclocked. But it will crash Furmark/OCCT stock. Heaven also seems fine, although I only tested this for about 30 min, still better than the 5 min lasted on the other stress tests.
 
Last edited:

MrK6

Diamond Member
Aug 9, 2004
4,458
4
81
I have the Asus 7950 DC II which is a 3 slot cooler. My PSU is a Corsair TX650 which has Seasonic internals and should be more than enough for a single cars setup.

Does this sound like a faulty card or poorly designed card in general? Really considering returing it. Like I said, I can play Skyrim for hours and my temps wont breach 65C with an idle of 30C and this is while overclocked. But it will crash Furmark/OCCT stock. Heaven also seems fine, although I only tested this for about 30 min, still better than the 5 min lasted on the other stress tests.
Like I said, because the VRM's probably aren't that heavily cooled, and their heatsinks are separate from the main GPU heatsink. As long as you don't run power virus programs or pump crazy voltage through the GPU for an overclock, youll be fine. Install GPUTweak by ASUS and check your VRM temps in a game and in Furmark. You'll see the difference.
 

fstime

Diamond Member
Jan 18, 2004
4,382
5
81
Is this problem more or less universal with all 7950's or just a few designs? How would increasing GPU voltage put the card in danger, do you mean don't increase the memory voltage?

Does your 7970 have the same issue?

My highest stable o/c is 1050/1650 (tested by gaming). Oddly, the ASUS stock voltage is .993, not the reference spec 1.093. The o/c was done with on 1.093V.

Thanks for your help.
 
Last edited:

skipsneeky2

Diamond Member
May 21, 2011
5,035
1
71
Like I said, because the VRM's probably aren't that heavily cooled, and their heatsinks are separate from the main GPU heatsink. As long as you don't run power virus programs or pump crazy voltage through the GPU for an overclock, youll be fine. Install GPUTweak by ASUS and check your VRM temps in a game and in Furmark. You'll see the difference.

Nice oc on the 7970,the highest i have seen of any 7970 owner in the forums.

Must share your core voltage:thumbsup:,and from what i gathered 1700mhz is typical of any 7970 right?

Hate sounding like a n00b but still gathering oc info as we speak to get that maximum oc...i'm already at 1200/1650mhz at 1.186v.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,570
10,202
126
Not really true. I believe that by design, GPU's are easier to exploit via a power virus, so their loads are multitudes higher than would be on a CPU. Furthermore, a video card's PCB is much smaller than that of a motherboard, so they make do with what space they have.
Some 7950 models have the bare minimum cooling on the VRM's. While it's more than adequate for stock settings, a power virus will exploit this deficiency and crash the card due to the high temps. Don't run Furmark or similar programs anymore as you will eventually damage your hardware.
"Power virus" is a total misnomer, and is a made-up term invented to disguise poor engineering practices.

GPUs have many processors, and just like any processor, should, in theory, be able to be 100% utilized. But like many things, they are engineered only to handle a certain percentage of total load, rather than the full extent of total load.

Edit: Imagine if bridges were designed, only to handle medium or light traffic, and not a full highway full of cars. Would that be acceptable to you? Like I said, GPUs are under-engineered.

Edit: Imagine if CPUs were designed this way, and the engineer simply chose to label things like Linpack (which is a mathematical matrix solver or something like that), as a "CPU power virus", and said that running such programs was unsupported on those CPUs. Would that be acceptable to you, that only certain software codes were "supported", and others were not?

To me, for a general-purpose computing device, to be told that certain classes of programs were unacceptable to run on that hardware, due to causing high power-consumption within that processing device, would be totally unacceptable to me to use or purchase.

A properly-engineered, general-purpose computing device, should be able to run ANY program that is written for it, successfully, as long as the software doesn't contain programming defects. Causing the underlying hardware to draw too much power is not a software defect.
 
Last edited:

MrK6

Diamond Member
Aug 9, 2004
4,458
4
81
Is this problem more or less universal with all 7950's or just a few designs? How would increasing GPU voltage put the card in danger, do you mean don't increase the memory voltage?
The 7950 card designs were left to the board partners, so each company has its own design that "follows" AMD's specifications. Because of this, there's a lot of variance in the designs and how robust some are. Some 7950's have really beefy cooling all around, some are hit and miss.

The take home message is don't increase any voltages until you can check your VRM temps. Because you have an ASUS card, their GPUTweak program can easily read the VRM temps. So install it and check out your temps under load: they should be <100C at all times. Once you know your VRM temps, you can then move on to overclocking.

Does your 7970 have the same issue?
No, but I custom cooled mine. I also have a reference 7970, and the stock cooling solution is pretty beefy as well.

My highest stable o/c is 1050/1650 (tested by gaming). Oddly, the ASUS stock voltage is .993, not the reference spec 1.093. The o/c was done with on 1.093V.

Thanks for your help.
Like I said, check your VRM temps and then proceed with the oveclocking. It's always better to get more information :thumbsup:.
 

MrK6

Diamond Member
Aug 9, 2004
4,458
4
81
Nice oc on the 7970,the highest i have seen of any 7970 owner in the forums.

Must share your core voltage:thumbsup:,and from what i gathered 1700mhz is typical of any 7970 right?

Hate sounding like a n00b but still gathering oc info as we speak to get that maximum oc...i'm already at 1200/1650mhz at 1.186v.
It actually goes higher, but that's the highest stable I can get through a quick OC on in Afterburner @ 1.3V (1.2V with droop). If I flash it to a better BIOS, I can hit 1400MHz+. My RAM actually only hits 1600MHz stable. I have to give it 1.7V (stock is 1.6V) to hit 1775MHz stably. However, my core is water cooled.
"Power virus" is a total misnomer, and is a made-up term invented to disguise poor engineering practices.

GPUs have many processors, and just like any processor, should, in theory, be able to be 100% utilized. But like many things, they are engineered only to handle a certain percentage of total load, rather than the full extent of total load.
That's not how it works. GPU's and other microprocessors are designed with certain specifications, much like anything that is engineered. Power viruses and similar programs are designed not to provide functional work, but to simply suck as much current as possible. No company would waste time over-designing hardware so a few geeks can get excited knowing their GPU's can pull 400W and still function. That's a waste of resources on the company's part and as a consumer I certainly don't want to pay for it. It is therefore cheaper and easier to implement OCP (both hardware and software) to make sure the GPU is running as designed.

Edit: Imagine if bridges were designed, only to handle medium or light traffic, and not a full highway full of cars. Would that be acceptable to you? Like I said, GPUs are under-engineered.
You either made a poor analogy or don't understand how bridges are made. Just like I said above, anything engineered is made to specifications. You'll notice many smaller bridges will have a "tonnage limit" sign with pictures of certain semi-trailer trucks on it. These bridges are constructed with the intent to handle local traffic, and larger trucks will find alternate, mainstream routes. Just the same, large highway bridges too are designed to specifications, only there's are designed to handle the highway full of vehicles.

Edit: Imagine if CPUs were designed this way, and the engineer simply chose to label things like Linpack (which is a mathematical matrix solver or something like that), as a "CPU power virus", and said that running such programs was unsupported on those CPUs. Would that be acceptable to you, that only certain software codes were "supported", and others were not?

To me, for a general-purpose computing device, to be told that certain classes of programs were unacceptable to run on that hardware, due to causing high power-consumption within that processing device, would be totally unacceptable to me to use or purchase.

A properly-engineered, general-purpose computing device, should be able to run ANY program that is written for it, successfully, as long as the software doesn't contain programming defects. Causing the underlying hardware to draw too much power is not a software defect.
Again, you're in the minority of people who actually care and furthermore would be willing to pay the sizeable overhead this would cost. Some AIB's do make more robust graphics cards with more power phases and beefier components, but they generally are much more expensive.