is furmark safe?

[GMcDougal]

i want to stress my 4890 out, and see what kind of oc's i can get out of it without artifacting. is furmark safe? ive read its too harsh on graphics cards. can anyone recommend a program thats safe and stresses out the card if furmark isnt safe?

 

[Elledan]

It is pretty harsh and there have been some issues with some cards where apparently the VRMs/cooling solution wasn't completely ready for truly 100% usage. Then again, any decent GPU should be designed for 100% load, even if it isn't 24/7, especially if it can do OCing as well.

It's a good way to find out the OCing potential of your card at least You can always RMA the card if it fails during Furmark as this would indicate a design/manufacturing flaw.

 

[PrincessFrosty]

There is no way for an application to stress your card so that it damages it under normal operation. Furmark simply makes use of as much graphics processing power as it can and as such causes almost exactly 100% load on the GPU constantly.

If you damage a card through use of furmark and it's not been modified (i.e not overclocked) then you need to RMA the card for a replacement because the card isn't functioning correctly.

If you overclock your card then furmark is the best test to see if the OC's are stable, what is stable in say 3dmark might not survive a furmark test. In fact if you're overclocking your card you should really only be using furmark as the stress test, you need to ensure your clocks are stable.

Make note that overclocking voids your warranty and if you overclock and test the card in furmark and it happens to break or fail (which is unlikely as long as you know what you're doing) then you shouldn't RMA for a new one.

 

[Copyright]

There is no way for an application to stress your card so that it damages it under normal operation. Furmark simply makes use of as much graphics processing power as it can and as such causes almost exactly 100% load on the GPU constantly.

If you damage a card through use of furmark and it's not been modified (i.e not overclocked) then you need to RMA the card for a replacement because the card isn't functioning correctly.

If you overclock your card then furmark is the best test to see if the OC's are stable, what is stable in say 3dmark might not survive a furmark test. In fact if you're overclocking your card you should really only be using furmark as the stress test, you need to ensure your clocks are stable.

Make note that overclocking voids your warranty and if you overclock and test the card in furmark and it happens to break or fail (which is unlikely as long as you know what you're doing) then you shouldn't RMA for a new one.

Overclocking doesnt void your warranty... EVGA , XFX and some others are just fine with it.

 

[hhara]

I pushed one of my GTX 280's to 102C with Furmark before stopping. Fun stuff!

So yea, it is pretty damn harsh. If you really want to stress the video card then I think Furmark is your best bet, since any other program wont stress it as much. If it gets too bad for your liking, just stop it.

 

[xeddex]

Then again, any decent GPU should be designed for 100% load, even if it isn't 24/7, especially if it can do OCing as well.

I disagree. GPU's are not designed to run benchmarks; they're designed to play games etc. If no real world games comes close to causing all 100% of a gpu to be used then there is no reason to design a gpu that way. Sure I guess they could but at what cost? How many people would benefit from a gpu that was happy with 100% usage if they never run anything that uses anywhere near 100%?

All the gpu torture tests that artificially push a gpu in completely unrealistic ways to achieve 100% usage are just fine for epeen contests but matter exactly zip when it comes to running a "normal" program.

<gross generalization>
That's like NASA designing a rocket that can go to Mars and then only ever using it to go to the moon. If the job is to go to the moon then that's the job, end of story. If the job is to go to Mars then that's a different subject.
</gross generalization>

 

[Nihilanth99]

Furmark + R7XX is just not a good combination. The boards aren't designed for those kinds of current draw, the type only really seen in Furmark and OCCT. As far as I know, Catalyst will gimp the R7XX cards when Furmark is detected to precent the overcurrent protection from kicking in, and Cypress boards are supposed to downclock themselves automatically when such a situation comes up before the overcurrent protection shuts the card down.

As others have said, you could consider it a lack of robustness, but AMD/ATI must have seen it as an area worth sacrificing given the low odds of the board ever seeing such a situation.

 

[jeremyshaw]

I disagree. GPU's are not designed to run benchmarks; they're designed to play games etc. If no real world games comes close to causing all 100% of a gpu to be used then there is no reason to design a gpu that way. Sure I guess they could but at what cost? How many people would benefit from a gpu that was happy with 100% usage if they never run anything that uses anywhere near 100%?

All the gpu torture tests that artificially push a gpu in completely unrealistic ways to achieve 100% usage are just fine for epeen contests but matter exactly zip when it comes to running a "normal" program.

<gross generalization>
That's like NASA designing a rocket that can go to Mars and then only ever using it to go to the moon. If the job is to go to the moon then that's the job, end of story. If the job is to go to Mars then that's a different subject.
</gross generalization>

There will be games that use 100% of the GPU, there will be applications that use 100% of the CPU.

Why develop it otherwise?


Why do we have cars that can go 180MPH, and speed limits of 70MPH?

If the thing cannot go to 100%, then it's faulty. No matter if any so called 'real world' applications do not use 100%, as there are certan 'real world' applications which can.

<clean>
NASA designs a rocket to go to the moon, they aren't going to develop it to only goto the moon and back, they will design it with a little margin for error. While that margin will NOT get you to Mars, it will allow minor craptacular to happen, and still get back home
</clean>

 

[Exeodus]

Well, I read on Anandtech that ATI is not a fan of Furmark, they believe it is too hard on the card. All I use to test stability is a couple of 3Dmark runs, and 6 loops with the Crysis Warhead bench. If it passes that, you should be good.

 

[xeddex]

There will be games that use 100% of the GPU, there will be applications that use 100% of the CPU.

What games? What applications? I don't know of anything that will peg a gpu at 100% other than a couple of benchmark programs (furmark being one of them).


Why develop it otherwise?

Why do we have cars that can go 180MPH, and speed limits of 70MPH?

You're kidding right? The reason why there are cars that can go 180mph and speed limits of 70mph is irl-epeen. No more, no less.

If the thing cannot go to 100%, then it's faulty.

who says what "100%" is? If a gpu is designed to work at a specific level of maximum performance then THAT is 100%. Not what some damn benchmark says is "100%". Believe it or not electronics aren't supposed to have every one of their circuits maxed all at the same time. That's not a design "flaw".

 

[jeremyshaw]

If a device does not work to it's specifications...


then it's faulty.

Since it cannot work at 100%..... merely full capacity, not even counting OC's own addons....

then it's faulty.


Faulty is a device that doesn't work to it's stated/rated capacity.

What games? What applications? I don't know of anything that will peg a gpu at 100% other than a couple of benchmark programs (furmark being one of them).


Why develop it otherwise?




You're kidding right? The reason why there are cars that can go 180mph and speed limits of 70mph is irl-epeen. No more, no less.




who says what "100%" is? If a gpu is designed to work at a specific level of maximum performance then THAT is 100%. Not what some damn benchmark says is "100%". Believe it or not electronics aren't supposed to have every one of their circuits maxed all at the same time. That's not a design "flaw".

Quoted for posterity, and seriously...

100% is when everything is being used at the same time...
It is a design flaw when it fails to work.

Besides, you are assuming gaming is the only applications for graphics cards (or as nVidia loves to call them: CUDA cards).

 

[PrincessFrosty]

Overclocking doesnt void your warranty... EVGA , XFX and some others are just fine with it.

Yes it does with almost all cards, these business are an exception to the rule.

I disagree. GPU's are not designed to run benchmarks; they're designed to play games etc.

No, GPUs are designed to render images from code, that includes benchmarks.

You cannot produce a piece of hardware which starts to fail when something uses it to it's full potential, if 100% of it's current speeds is too high then they need to lower the speeds so that it functions correctly under any circumstances.

Furmark + R7XX is just not a good combination. The boards aren't designed for those kinds of current draw, the type only really seen in Furmark and OCCT. As far as I know, Catalyst will gimp the R7XX cards when Furmark is detected to precent the overcurrent protection from kicking in, and Cypress boards are supposed to downclock themselves automatically when such a situation comes up before the overcurrent protection shuts the card down.

As others have said, you could consider it a lack of robustness, but AMD/ATI must have seen it as an area worth sacrificing given the low odds of the board ever seeing such a situation.

This is interesting, do you have any links to additional information on AMD throttling in the presence of Furmark?

Cypress throttles the core down when 100 degrees temperature is reached, it's a safety measure that amost all GPU/CPUs have now a days, we should be thankful it exists to stop people from frying cards.

I personally dont think the trade off between instability and that few extra Mhz is really worth it, you can dial your overclocks back a few Mhz and Furmark will work fine and you'e not going to notice that in games, but you're then covered for every eventuality.

 

[Nihilanth99]

This is interesting, do you have any links to additional information on AMD throttling in the presence of Furmark?

http://blogs.zdnet.com/hardware/?p=2456

 

[mafa]

if you compare the furmark results from both camps. The ATI cards have significant higher scores than nvidia cards, even a 4850 produces a higher score than gtx 285.

So furmark either over-stress ATI card or not putting enough stress on the nvidia card. You can't blame ati for "throttling".

 

[Copyright]

Yes it does with almost all cards, these business are an exception to the rule.



No, GPUs are designed to render images from code, that includes benchmarks.

You cannot produce a piece of hardware which starts to fail when something uses it to it's full potential, if 100% of it's current speeds is too high then they need to lower the speeds so that it functions correctly under any circumstances.



This is interesting, do you have any links to additional information on AMD throttling in the presence of Furmark?

Cypress throttles the core down when 100 degrees temperature is reached, it's a safety measure that amost all GPU/CPUs have now a days, we should be thankful it exists to stop people from frying cards.

I personally dont think the trade off between instability and that few extra Mhz is really worth it, you can dial your overclocks back a few Mhz and Furmark will work fine and you'e not going to notice that in games, but you're then covered for every eventuality.

Pretty sure Asus and MSI with their overclocking utilities are not going to void your warranty. EVGA and XFX I know do not and there have been many threads about that. As long as there is know physical modification to the card to allow for higher overclocks I highly doubt any company would even know if you did.

 

[SicKlown42012]

Furmark with an ATI 4800/5800 is safe, mainly because AMD's drivers will detect if the program is running(either by name or type of load) and throttle the card to prevent damage. Google "AMD Power Virus" to get a handful of articles that explain the situation and AMD's reasoning for doing this.

 

[Blkout]

i want to stress my 4890 out, and see what kind of oc's i can get out of it without artifacting. is furmark safe? ive read its too harsh on graphics cards. can anyone recommend a program thats safe and stresses out the card if furmark isnt safe?

Just run some loops in 3DMARK06 or Crysis Warhead, it's as stressful as your video card is ever going to see. If your overclock is stable in those apps without artifacting, locking up, or rebooting, then you're just fine.

 

[defaultluser]

Furmark + R7XX is just not a good combination. The boards aren't designed for those kinds of current draw, the type only really seen in Furmark and OCCT. As far as I know, Catalyst will gimp the R7XX cards when Furmark is detected to precent the overcurrent protection from kicking in, and Cypress boards are supposed to downclock themselves automatically when such a situation comes up before the overcurrent protection shuts the card down.

The Cypress method is the same method Intel uses when they detect the processor is overheated: the core downclocks until it falls below the temperature threshold.

This is a perfectly reasonable approach, assuming your processor doesn't typically create more heat than the cooling solution can handle. Intel got into trouble with this on some of the top-end Prescott P4s, but they have since dumped the P4, and never had that problem again.

Intel's solution to this problem was to set a wall at 95w TDP for consumer-level chips, which means even a below-spec cooling system won't leave the processor constantly throttling. For the processors with higher power consumption, they've introduced a more aggressive technique to take full advantage of the TDP ceiling called Turbo Boost.

But video cards are the realm of 100+w workloads, so I think it will be interesting to see if the safety checks are too aggressive. If so, users could start to see their cards throttle during a normal game!

 

[PrincessFrosty]

http://blogs.zdnet.com/hardware/?p=2456

Thanks.

Pretty sure Asus and MSI with their overclocking utilities are not going to void your warranty. EVGA and XFX I know do not and there have been many threads about that. As long as there is know physical modification to the card to allow for higher overclocks I highly doubt any company would even know if you did.

EVGA and XFX have specific alterations to their warranty that allow overclocking of the card, unless specifically stated overclocking is allowed then warranry wont cover it, it's baically operating the card outside of its normal use.

it's fairly hard to bin a card now a days due to overclocking alone, near impossible I would say. Overvolting a bit more risky I would say. Most of the damage now a days comes from people running high overclocks for sustained periods of time causing components to have much shorter lifespans which for a lot of business exceeds the warranty time anyway so isn't their problem (bar XFX i think)

 

[Nihilanth99]

The Cypress method is the same method Intel uses when they detect the processor is overheated: the core downclocks until it falls below the temperature threshold.

That's true with Intel, but as far as I know, it was a current problem with the 4890, not a temperature problem.

 

[BababooeyHTJ]

There is no way for an application to stress your card so that it damages it under normal operation. Furmark simply makes use of as much graphics processing power as it can and as such causes almost exactly 100% load on the GPU constantly.

If you damage a card through use of furmark and it's not been modified (i.e not overclocked) then you need to RMA the card for a replacement because the card isn't functioning correctly.

If you overclock your card then furmark is the best test to see if the OC's are stable, what is stable in say 3dmark might not survive a furmark test. In fact if you're overclocking your card you should really only be using furmark as the stress test, you need to ensure your clocks are stable.

Make note that overclocking voids your warranty and if you overclock and test the card in furmark and it happens to break or fail (which is unlikely as long as you know what you're doing) then you shouldn't RMA for a new one.

You do know that OCCT's GPU test will make any reference 4870/90/X2 shut down. I think that OCP kicks in. Try it out full screen, test 8 (I think it is), while forcing no post processing via CCC. It won't take long.

Also at leat with the early 4870s quite a few were taken out with Furmark. On others the memory was degraded. There is a reason that you had to rename the exe on ATI cards to get furmark to run at full load. You might still have to rename the exe, I'm not sure.

 

[PrincessFrosty]

Yeah obviously there was a defect in the 48xx design where temperature throttle wasn't kicking in correctly, even if you could fry a card under normal circumstances you'd be eligable for an RMA.