Why are people generally so rigorous about CPU stability when OC'ing, but so lax when it comes to GPU overclocking?

[dug777]

... as creating 'excessive' or 'unrealistically high temperatures' that you will not experience in real gaming.

As far as I'm concerned, my GPU isn't stable unless it can manage an overnight furmark stability test.

If it can't, then it's unstable and as such any conclusions i draw as to the stability of my drivers or anything else are invalid. You have to have complete control all the variables except the one you are testing.

I will also beg to differ regarding the belief that furmark creates unrealistically high temperatures, on two levels.

1) Most people don't have dual monitors and as such can't monitor their temperatures in real time (GPU-Z does this, if you wish to). Unless you have done this, you don't actually know how hot your card gets while gaming. Temperatures drop increadibly quickly and alt-tabbing out will not give you a true understanding of this.

2) Games such as Farcry and Serious Sam 2 (using two examples that I am personally familar with, I am sure there are others) will cause my 4850 to attain the same temperatures as furmark.

I generally find that newer games, such as Crysis and COD4 to name two i am personally familar with and have tested, will not attain the same temperatures or power draw (monitored using a KillaWatt). I assume this is because they stress the core less evenly, being bottlenecked in a particular way (almost certainly memory for my 4850).

Anyway, I hope this invites some discussion on the matter. I am happy to provide further detail if any has any questions.

 

[Ares202]

i get a constant 99% GPU usage on GRID even when in menu, thats the same as what furmark does, after about two hours of playing it goes 10c higher than playing for example crysis or call of duty 4 where the usage drops in certain points

 

[taltamir]

what is the WORST that can happen if the CPU makes an error? a corrupt OS file requires you to reinstall windows, data gets corrupted writing to the disk and you lose and important file. the system blue screens, etc...

What is the WORST that can happen if the GPU makes an error? vista restarts the video adapter and you get the message (Display driver atikmdag/nvlddmkm stopped responding and has recovered) while your video game crashes... SOME video games can reinitialize their video with the rebooted video card (so all you have is a few seconds of black screen).
And sometimes its just gonna be a single artifact on the screen showed for a fraction of a second.

Overall the GPU stability matters a lot less... On the other hand, if you use it for computation, like folding, it has to be rock solid.

 

[dug777]

Originally posted by: taltamir
what is the WORST that can happen if the CPU makes an error? a corrupt OS file requires you to reinstall windows, data gets corrupted writing to the disk and you lose and important file. the system blue screens, etc...

What is the WORST that can happen if the GPU makes an error? vista restarts the video adapter and you get the message (Display driver atikmdag/nvlddmkm stopped responding and has recovered) while your video game crashes... SOME video games can reinitialize their video with the rebooted video card (so all you have is a few seconds of black screen).
And sometimes its just gonna be a single artifact on the screen showed for a fraction of a second.

Overall the GPU stability matters a lot less... On the other hand, if you use it for computation, like folding, it has to be rock solid.

Good points :beer:

Games crashing annoy me to the point that I'm happy to eliminate oc'ing related heat as a culprit, after all it's not particularly hard to do.

 

[Tempered81]

for graphics cards, i just oc to the point where my games crash, and back it down 1 mhz

 

[SlowSpyder]

I'm pretty lazy about both actually. I'll run some stability tests, like prime for an hour and 3DMark a few times. If that passes I just use it, if games crash I'll revisit my overclocks. If the PC just works I'm happy. But, I never go for every last MHz either. Take my 4870 for example, it's been rock solid at 820MHz, I could install Rivatuner to go for more, maybe I'd get another 10-20MHz, but I don't care. Lazy I guess...

 

[yh125d]

Just run HWmonitor for temp monitoring, play for 20 minutes, and look at the max temp on the GPU core for its load temps. Or nvidia's monitor, or any other temp tool that has a graph. Dual monitor for temp readings unneeded

 

[AzN]

Why do you need to test it over night? A good 30 minutes should test for stability until the temperatures are at the highest peak and stays there. When there's no errors you wouldn't have problems running it over night either.

Personally use ATITOOL to test it but in a real game scenario the vid card doesn't get as hot.

 

[AzN]

Originally posted by: taltamir
what is the WORST that can happen if the CPU makes an error? a corrupt OS file requires you to reinstall windows, data gets corrupted writing to the disk and you lose and important file. the system blue screens, etc...

What is the WORST that can happen if the GPU makes an error? vista restarts the video adapter and you get the message (Display driver atikmdag/nvlddmkm stopped responding and has recovered) while your video game crashes... SOME video games can reinitialize their video with the rebooted video card (so all you have is a few seconds of black screen).
And sometimes its just gonna be a single artifact on the screen showed for a fraction of a second.

Overall the GPU stability matters a lot less... On the other hand, if you use it for computation, like folding, it has to be rock solid.

If you are using vista you can simply restore all your system files.

 

[Tempered81]

Originally posted by: yh125d
Just run HWmonitor for temp monitoring, play for 20 minutes, and look at the max temp on the GPU core for its load temps. Or nvidia's monitor, or any other temp tool that has a graph. Dual monitor for temp readings unneeded

Yah, even gpu-z has a sensor tab with graphed temps... you can go play for 10-15 minutes and heat up the vmem & gpu, then close the game come back to windows and see what your old temps were.

 

[dguy6789]

Personally, I can't stand using a machine that will crash under any circumstance. If there is a way to make the machine crash that isn't due to a software bug, then something needs to be turned down. I would never run my pc at settings that it could not pass any benchmark indefinitely, cpu, gpu, memory, whatever.

 

[MyLeftNut]

My current 8800GT when OC'd and unstable will crash my entire system when just browsing websites. It happens maybe only twice a week, but ever since I backed the OC down, like way down, it hasn't crashed for 2 months.

 

[Lonyo]

What about rthdribl? Plus (assuming things play nice) you can run it in a window and monitor GPU temps at the same time as it's running.
I found it usually manages to heat up my GPU pretty effectively, although I haven't compared it to games.

 

[aka1nas]

Most consumer GPU's simply won't run 24/7 stress tests without errors, period. The GPU vendors have gotten away with it because performance is what sells and until relatively recently there wasn't much of an issue with the occasional calculation error.

I would imagine that part of the additional cost of the HPC solutions like Tesla are for the extra validation and testing to ensure they can run stably for long periods under load.

 

[imported_Scoop]

I OC to the point that atitool won't give me artifacts. Haven't had stability problems due to my GPU.

 

[PingSpike]

Its what taltamir said....I'm sure you can get the PC to hard crash with an overclocked video card, but I'd be more inclined to think it had more to do with an insufficient power supply for the overall system power then the video card itself. An overclocked OS, bus, ram, etc has the potential to corrupt files or at least really screw up your PC. An unstable video card, while annoying, has much less potential to screw you over.

 

[AzN]

I run my computer 24/7 with it overclocked for 2.5 years now. Not a single hickup. I don't even reboot until a patch or something.

I've got pretty heavy overclocking on my CPU, bus, Ram, GPU. If you have crappy components you are bound for a crash. If you have quality stuff you shouldn't get a crash. Testing for stability is mandatory. Orthos, ATITOOL, MEMTEST. Once you pass these 3 tests your computer shouldn't get a crash.

I get a crash every now and then on a new released game usually contributed to a bugs when everyone else is having the same problem or some crappy software that is just buggy period. Other than that I don't get a crash due to my overclocking. I've had many days of 6-8 hours 1 sitting gaming sessions. NEVER FAILED ME. IF IT DID I WOULD FIX IT!!!

 

[OCNewbie]

What is the latest version of ATITool btw?

 

[chizow]

I'm always more conservative and thorough with CPU overclocks because any instability is harder to detect, reproduce and diagnose. Unstable GPU overclocks are easily diagnosed and isolated from the rest of the system however, and can often be reliably repeated or exposed. If your CPU is unstable it can manifest itself in numerous ways and may end up leading to troubleshooting measures that are completely unnecessary.