Why do some games crash with an OC?

[SK-1]

Just always wondered. I notice this with some titles not all. BF4 and Titanfall come to mind. I actually downclock my card to help stop crashes.

 

[...PACMAN...]

If they crash with an overclock, it's not a stable overclock. Different engines can cause more stress, you need an overclock that is stable across the board.

 

[Toothless]

2fast4u

Unstable overclocks as @...PACMAN... said.

 

[SK-1]

So my factory overclocked graphics card was shipped unstable? And why only certain games...My CPU OC is as stable as it gets...

 

[m6tzg6r]

Possibly, i had an ASUS that was unstable out of the box before.

 

[Toothless]

So my factory overclocked graphics card was shipped unstable? And why only certain games...My CPU OC is as stable as it gets...

When did you get it? Over time GPUs degrade and so overclocks can become unstable after a while.

 

[Cvrk]

I actually downclock my card to help stop crashes.

Just Frostbite Engine. Same problem with Dragon Age Inquisition.
I searched the internet forever to solve my problem back in january 2015. I played DA on a friends pc,and he had an Asus card R9 OC . Searched for days on end , finally i found the answer on the Bioware forums. It crashed every 10 minutes. Downclocked with just only 10MHz i think (1010 to 1000 gpu frequency), and never had a problem .

 

[SK-1]

Just Frostbite Engine. Same problem with Dragon Age Inquisition.
I searched the internet forever to solve my problem back in january 2015. I played DA on a friends pc,and he had an Asus card R9 OC . Searched for days on end , finally i found the answer on the Bioware forums. It crashed every 10 minutes. Downclocked with just only 10MHz i think (1010 to 1000 gpu frequency), and never had a problem .

Thanks for the input. I did a little due diligence (other forums, ect.) and found the same. Basically OC sensitive software. Even a little oc messed with some of these games. My oc experience spans well over a decade and a half so I was confident it was most likely sensitive software. I've seen this issue persist over many platforms.

 

[Frag_Maniac]

My Sapphire 7970 OC easily artifacts on some games like Far Cry 3 if I OC it at all, but it doesn't artifact or crash with the factory OC. If yours does, that is unacceptable and cause for RMA.

Curious, what is your ASIC Quality reading? If you don't know how to check it, launch GPU-Z, click the little green icon in the far upper left corner, and click Read ASIC quality.

Mine is only 61%, and mostly why I think it won't OC on air well, but some have said it's not necessarily that, despite W1zzard indicating it's typical in his chart on the tool.

I've been told a low ASIC reading is not a legit reason to RMA a card, but I would think if it can't even run stable at the factory OC, regardless what the game is, that should be a legit reason.

Is this regarding the Giga Windforce 770 in your spec chart? If so, have you checked to see if all it's fans are running, and if the temps are normal?

 

[Cvrk]

Regarding what @FragManiac said . He makes a point.
However it all depends on the country you live.
I almost 90% can not send back a product after purchase. They got a 3 day return money policy that works well. After that nothing.
What is the cause of your return ? - I have very low ASIC, or i crash in some game all the time etc . Nobody cares. Unless it has physical damage or artifacts in every 3d application that 100% proves the card is broken.
You power up the computer, do you have video image ? Yes....well then that's it, the card is perfect. This is how things work in some places of the world.
Lowering the freq with 10 Mhz so you can play a game is a very easy fix. Trying to get refund ,you will hate your life.

 

[Vayra86]

There is a correlation between temperature, voltage and the amount of leakage you will incur on a chip. This is also the reason Nvidia's boost clocks are self-binning and will go lower when temps reach a certain threshold. The GPU can get hotter and still work, but the amount of current leakage will increase, thus making that performance unstable. Boost clocks are in a way, a very gradual kind of throttling, except this throttling is voltage related instead of temperature limited.

Heavier engines tax the GPU and increase temperature > GPU will get forced into max boost / max vCore > higher temperatures increase current leakage > you get crashes. That is usually how it works. This is why we try to stress overclocks to 'a little more' than a typical gaming scenario.

Just having 100% load on a GPU does NOT mean all the internal components of the chip are working at full power, it just means SOME part of the GPU is fully stressed. An example: the large amounts of Tesselation used in the Heaven benchmark as opposed to the limited Tesselation in the Valley benchmark: many overclocks that are 'on the edge' will flawlessly run Valley but crash on Heaven.

This could be shaders, could be memory, or anything else that just won't run faster because the application saturates it. But then suddenly game engines improve and manage to put full load on several parts of the GPU. Suddenly the voltage required for 100% load will be slightly increased because more parts of the GPU are fully stressed, causing your overclock to become unstable.

 

[qubit]

The era clock speed sensitive software is long gone. Just think of all the different speeds and models of hardware out there and stuff works fine.

Try loading the overclock on your card to reference model speeds. If the problem goes away, then the card is faulty and you have a legitimate reason for an RMA.

 

[xfia]

just curious if you have a bunch of different vendor software running the show? specifically overclocking and monitoring. i hate all that stuff.. bunch of buggy crap. bios or nothing for me. got some nice fans and dont even bother checking my temps anymore.

 

[Mussels]

i just wanted to come here and see the answers... but yeah, people got it right straight away this time.

WAY too many people think an overclock is stable when its nothing of the sort. As an example that got me a few years ago, some games heated up my system more than others and my northbridge and CPU VRM's were linked together with a heatpipe. eventually the system would crash after about 4+ hours of gaming in certain titles, which was really hard to track down.

That heatpipe was hitting 90C and cooking one of my ram modules which was right next to it, causing corruption. Memtest didn't show anything because memtest didn't heat up the VRM's, meaning the ram was cooler and stable.

 

[xfia]

thanks mussels.. not always so simple when your crashing and cant figure it out. may think its your gpu and your cpu overclock is stable but it could be the other way around or not getting air moving over vrm's.
i would say to take down the overclock and go from there because if you rma a gpu that has nothing wrong with it you will get the same one back and waist your time.
edit-looks like a nice psu being run way under the thresh hold but how old is it? not unheard of to start rippling or undervolting.

 

[Aquinus]

The funny thing is that people don't realize that a GPU is not a monolithic structure. There are a lot of internal parts to it, much like a multi-core processor. What can happen is if the appropriate kinds of 3D API calls get made, it will put extra stress on certain parts of the GPU. It's like having regular CPU application that's really heavy on floating point math. You're going to stress the FPU out but, you very well might never know if the integer cores are unstable at full tilt. So when you use certain engines, it will sufficently fill up the GPU's caches and will put more stress on more parts of the GPU. So you might have two games that are using "100% of your GPU," really just means every clock cycle is being used. It doesn't say how much of the GPU is getting used per clock cycle.

So in summary, it's probably just the game engine taxing the GPU in a way that puts strain on the "weakest link" inside of the GPU so to speak. Every part of the GPU doesn't necessarily run in lock step.

This is the same reason why Furmark tends to hammer GPUs more than anything else. 3D calls in it are optimized in a way to do the maximum amount of GPU work and in turn, increased stress and power consumption.

 

[Mussels]

The funny thing is that people don't realize that a GPU is not a monolithic structure. There are a lot of internal parts to it, much like a multi-core processor. What can happen is if the appropriate kinds of 3D API calls get made, it will put extra stress on certain parts of the GPU. It's like having regular CPU application that's really heavy on floating point math. You're going to stress the FPU out but, you very well might never know if the integer cores are unstable at full tilt. So when you use certain engines, it will sufficently fill up the GPU's caches and will put more stress on more parts of the GPU. So you might have two games that are using "100% of your GPU," really just means every clock cycle is being used. It doesn't say how much of the GPU is getting used per clock cycle.

So in summary, it's probably just the game engine taxing the GPU in a way that puts strain on the "weakest link" inside of the GPU so to speak. Every part of the GPU doesn't necessarily run in lock step.

This is the same reason why Furmark tends to hammer GPUs more than anything else. 3D calls in it are optimized in a way to do the maximum amount of GPU work and in turn, increased stress and power consumption.

and of course the same thing applies to every part of your system as a whole - the ram, the northbridge, the IMC, your SATA controllers (if you FSB/BCLK OC'd) and so on, down to voltages and temperatures of the tiniest components.

 

[xfia]

microsoft recommends updating your audio driver if a game is crashing.. lol might seem strange but its apparently a pretty common problem. idk if any you mods work with all the troubleshooting tools microsoft uses but if i got a hard one i go to technet run what they tell me too.. stressing drivers and all that and they figure it out like nothing after i upload the files.

 

[SK-1]

just curious if you have a bunch of different vendor software running the show? specifically overclocking and monitoring. i hate all that stuff.. bunch of buggy crap. bios or nothing for me. got some nice fans and dont even bother checking my temps anymore.

Bullseye on the monitoring software. I never would have guessed. I've been overclocking longer than most TPU members have been out of the womb so I friggin knew my OC was as stable as say a "Shrimp on the barbie".... 100% crash free now! Thanks for the suggestion xfia...and to all who replied.

 

[natr0n]

Games like bf4 vary between high and low load/stress on different maps and such.
Nothing beats testing an OC with real games as they exploit any issues rather well.

 

[SomeOne99h]

Games like bf4 vary between high and low load/stress on different maps and such.
Nothing beats testing an OC with real games as they exploit any issues rather well.

Stress apps are like "Hardware Stressing Simulator".

 

[Jetster]

Might not be the over clock that is causing it. And you do realize that there is a bunch of old farts on TPU?

 

[SK-1]

Bullseye on the monitoring software. 100% crash free now! Thanks for the suggestion xfia...and to all who replied.

Might not be the over clock that is causing it. And you do realize that there is a bunch of old farts on TPU?

 

[xorbe]

You have to tickle the circuit that doesn't meet timing. Not all games run code that tickle the offending circuit.

 

[Flogger23m]

So my factory overclocked graphics card was shipped unstable? And why only certain games...My CPU OC is as stable as it gets...

Sounds like it. Or maybe your CPU is not as stable as you think.