Your favourite way of testing GPU stability

ktos2

n00b
Joined
Jul 2, 2016
Messages
13
Just like in the title, how do you guys like to test if your gpu oc is stable?
 
OCCT for about 30 minutes with error checking and then gaming with something demanding for an evening. I know that it results in lower overclocks to make sure its error-free, but its more stable.
 
Personally, I just fire up demanding games. It used to be Crysis, then Crysis 3, now it's usually AC Unity.
 
I usually use a a 3DMark stress test for a while and see if there are any crashes or artifacts. If everything looks ok there I fire up some games and see what happens.
 
Run Unigine in a few loops, Heaven and then Valley, then a few passes of 3dMark, then games, if you don't see snowflakes or crashes then you're good to go.
 
Fire up you game of choice, *the most stressful one you own preferably* and tweak from there. Tweaking an OC on a benchmark/Heaven to me after doing it so many times seems like a waste, more often than not it crashes once you get into a real game.
 
8 hours prime 95 (blend).
8 hours Linpack
8 hours memtest86

Games just don't test your CPU to the max nor do they show off small calculations errors.


You always will want to run some CPU+GPU stress benchmarks to make sure you PSU is up for it. its not much help to just stress thee CPU and find it stable. then lunch a game and fine you system is crashing because you PSU kan deliver or the cooling is to low.

So prime95 blend + furmark for that.


ive compiled a specific test suits for that here

www.techcenter.dk/Phoenix.exe
 
Heaven.

But Crysis 3 and Call of Duty Advanced Warfare (and now Rise of Tomb Raider too) have also been quick to detect unstable OCs.
 
Deus Ex Mankind Divided. Maxed out at 4K. LOL, even a Titan X might as well be a GTS 250 in that game.
 
Best is still games, especially if you mix between heavy and light games. For NV GPUs like Maxwell, a wildly fluctuating boost can cause issues even when your OC is stable at *full* load (like in benchmarks/stability tests).

For example, my highest OC (+180/+500, 1534/8000 effective) works with games that load the GPU fully, like W3, Dying Light, or DOOM (when not capped to 60fps) all day long. However, due to the rapidly varying load in Overwatch, that setting has a chance of crashing.
 
Unigen and Furmark for initial stressing; then gaming and running BOINC distributed computing applications (primarily Einstein @ Home in the long term, probably something with shorter tasks on a new card) on the GPU as continuous monitoring.
 
Nothing beats playing games with a very different engine and type of load. I've had OCs that won't budge on recent demanding games and benches but will crash in old games where the GPU load is 20%.

With CPUs though, if I pass torture tests I'm usually good to go. Sandy Bridge was an exception where super low loads or idle might give a x124 BSOD and require a bump of the vcore.
 
I play my games. If they don't crash and I don't see any artifacts, all the vitals like temp look under control, I'm happy.
 
In my experience, the surest way is to play various video games.

My GTX 1080 which is currently sent for RMA, had issues only in games that ran at high fps. Everything was fine when I ran 3DMark, but the moment I ran Path of Exile it artifact and crash. Didn't artifact in Serious Sam HD, but quickly crashed too.
Looked to me like it had a bad memory chip.

I'm not sure if there are some kind of error control or the compression mechanism is masking data corruption in certain situation, but it seems to me like whatever issue the card had only shows when rendering at high fps rate
 
I'm not sure if there are some kind of error control or the compression mechanism is masking data corruption in certain situation, but it seems to me like whatever issue the card had only shows when rendering at high fps rate

All gpus use memory compression and these latest generations use a brick ton of it. They actually mask the artifacts until it can't catch up and it takes a lot of speed to outstrip the error correction. It sounds like you did that with high fps games. But yours is a case of defect, not necessarily one of testing oc limits, still though it's the same process.
 
Back
Top