• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Do GPUs still throttle Furmark?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Culbrelai

Member
Joined
Oct 25, 2012
I heard GPUs (used to?) throttle this testing program so I went to test it.

Extreme burn in enabled... erm. Those are normal clocks for me (although how they stay synched with one running at ~1200 mhz and another at 1250 mhz is beyond me)

doesntseemlikeit_zps0b0ff1fc.png
 
I say not to use it because it can cause "dead GPU syndrome"
 
I can not see beating on your GPU like this ?

better off testing with a bench such as valley or heaven, or your favorite / most demanding game for that matter

TBH I am not a real big on Prime, Linx or Intel burn in

Stability defined by me means it will run what ever it is you want to run , be it games, productivity, multimedia or benching without stability issues
 
It's weird because I've seen it get the hotter 670 get to 100c gaming, and furmark only got it to 73.

Maybe they throttle it at a more hardware level now? hmm...
 
It's weird because I've seen it get the hotter 670 get to 100c gaming, and furmark only got it to 73.

Maybe they throttle it at a more hardware level now? hmm...

What do you mean "throttle it at a more hardware level"?
Throttling has always been hardware controlled.
 
What do you mean "throttle it at a more hardware level"?
Throttling has always been hardware controlled.

I thought it was drivers?

People used to say you could get around it by renaming Furmark's executable, which led me to believe it was software (drivers) maybe the original claim was always patently false, that the throttle was based on drivers.
 
Recent cards like Radeon HD 5000

LOL im dying.

Still, I suppose it makes sense considering most (read: all?) GPU monitoring programs like Afterburner only measure the core temperature of the GPU itself and not the various doohickeys and doo dads that supply power and such, and overstressing these can make them go blammo just as overly overclocking the cards could =P

And it kind of grinds my gears that the GPU makers would single out programs instead of providing better cooling solutions and beefier VRMs and such.

Which begs the question are there programs that do similar things to CPUs?
 
Last edited:
The latest version of P95 and Haswell CPU's, yes... sort of. It won't cause it to go blammo like Furmark can, however it will take things to another level versus previous versions. Some say not to use it even.
 
Those tools, especially Furmark, cause a unrealistic load with some uber level of stress. For example when driving a car always at the rev limit of maybe 10000 or whatever, the engine may fail after only a singe day of use, it is simply running "outside intended spec", a engine is not made to be used in such a way. A formula 1 engine at 19000 will only live about 12h or so when always driving at the rev limit, in comparison a Mercedes may life 30 years when always using the engine in a gentle way around 2000-3000 rev... so the lifetime of such limit-approaches is critically reduced (it can even be reduced to a single day comparable to a formula 1 engine) and same counts for nowadays high end processors.

The architecture of nowadays GPU and even CPU is made by using the resources in a way that the programs that are being used perform the best way possible. Those programs are having a unstable and acrobatic load and a lot of different mechanisms are running hidden to the user. But the load is never a "constant peak load", its not a rev meter always stuck at the max (Furmark and even P95 is acting like this), instead a game engine or other programs are very acrobatic in behavior and use different spots. But a rev program such as Furmark is hitting the same spot over and over and so incredible fast that the hardware will suffer critical load and may easely go boom because of constant overload. In comparison i guess no one is using their brand new Ferrari at the 1. gear and then pumping the rev to the max possible and driving from California to Florida all the way at a constant max rev... when they arrive the car may be boom... sorry for them. Because just as i said, a modern processor is tuned in such a way that it will always try to perform at the absolute max level, but it only is safe when a program is having a acrobatic and gentle behavior, for example a game engine. Or in other words, a program such as a game engine is using the gear switches properly, 1. gear then the 2. and 4. and maybe down to the 1. at some point... simply the way how a car engine should be used, and exactly the same is required for a modern processor. In term something is creating a uber stress or overload, the only safe approach is to insanely downclock the processors, maybe half of the "standart clock", kinda comparable when the rev of a car will be reduced to 6000 instead of 10000 or so, then it may be safe and thats the stuff the "throttling" is usually trying to apply, so its only natural in order to protect a hardware.

For stability test there could be many tools but i think best approach is simply to run a good benchmark program for 3h or so... some will offer a "loop mode" and in term it stays stable it can be considered "stable enough for gaming needs", i wouldnt OC stuff for other use anyway... so all what truly matters is if it can stay stable in gaming terms, and there is never a bulletproof stability needed, in fact a bulletproof stability doesnt exist, its a fairy tale. So it does not proof anything in term a CPU or GPU may have a stability failure after 6h of continuous stress testing stuff (some weird people are actually doing it and may have their stuff going boom after). Furmark is "rev limited" by drivers/hardware and it is perfectly fine because it should not be used at all, so the manufacturers are doing the right thing.

So in my mind, people should stop using the "crap", its no use and it may not even necessarily find a instability because not enough of acrobatics. Nowadays the architectures are designed in a way that they have so many different spots that can be used by high end software, that most "burning tools" simply cant hit those spots anymore, not versatile enough and doing nothing more than possible damage. Prime 95 may still be safe to use when a CPU is downclocked or at least not overclocked, for example 3 GHz or so instead of 4 GHz, but its surely not a program that should be used on overclocked CPUs. Overclocked CPUs have other prioritys that will work great running together with a game engine, but a "stress tool" will cause a totally unrealistic and sometimes damaging load on overclocked CPUs.

Again, best way of stability testing is simply running games or benchmarks... that will surely do the trick when watching properly.
 
Last edited:
Dude, no need for a novel on a thread where the answer is already clearly received...
 
Those tools, especially Furmark, cause a unrealistic load with some uber level of stress. For example when driving a car always at the rev limit of maybe 10000 or whatever, the engine may fail after only a singe day of use, it is simply running "outside intended spec", a engine is not made to be used in such a way. A formula 1 engine at 19000 will only live about 12h or so when always driving at the rev limit, in comparison a Mercedes may life 30 years when always using the engine in a gentle way around 2000-3000 rev... so the lifetime of such limit-approaches is critically reduced (it can even be reduced to a single day comparable to a formula 1 engine) and same counts for nowadays high end processors.

I think still its BS that high end graphics cards either cooling solutions and/or VRMs/mosfets etc can't handle it. Considering GTX 980s ARE Formula 1s in the graphics cards world.
 
My knowledge is actually pretty limited when it comes to extreme overclocking because no experience. It is useful to see the possible behavior of material under extreme condition. As far as i know under LN2 conditions at crazy binned clocks (~2000 MHz core clock or even more) there is so much power going through the silicon that the possible lifetime is not much more than a formula 1 engine. But what actually happens here? OC cards with powerful super -caps, -choke, -mosfet, -VRM and whatever are not going to have a issue here as long as cooled properly, not even when 800 W affecting a single chip. But no matter how well a chip cooled, the electro migration will be hard and the aging progress is probably speed up by 1000+ times. Doesnt take long for the chip to degenerate itself and max clock will slowly go down and at some point it cant handle the load and it may even break... irreversible. The loss, partially or even complete is always irreversible. The structure of the silicon can be damaged or reduced at many spots, so a chip is able to get killed by to much power, be it insane LN2 clocks or simply a crazy rev value... it all may have similiar results. Either some single spots inside the chip simply overheat (no matter the temperature at the die, its way to inaccurate), or the structure will die because the huge migration will simply burst it away and the structures is just a few nm... it is very sensitive and delicate. To much power can damage the chip structure, not a secret at all. But when the chip is cooled down the hard way, the electrical resistance will greatly be reduced, stability will increase and risk of damage is lowered big deal because much less migration. Although, 20 C or so doesnt make the true difference... 100 C lesser or below zero at least is giving true benefit.

Answering the question, well, the GTX 980 may look like a formula 1 car but in fact it is a very limited locked down chip with hard power target. On top of that it is very possible that the chip may not even be fully unlocked... i know Nvidia, they only offer as much as needed and will strike back with the true potential as soon as a Radeon is creating a threat. The new name will probably be "980 ti", or "Titan", who knows. It is very possible that a overloaded chip can be killed, not even the slightest doubt.

Dude, no need for a novel on a thread where the answer is already clearly received...
My "novel" is just a short explanation for inexperienced users, obviously not you, to be able to get the hang of it. The rev idea is actually from ED, and i think it is great as a comparison.
 
Last edited:
Back