How long should I stress test my GTX770 to ensure its 100% stable

[acvolta]

I recently decided to try and overclock my GTX770 (Gigabyte Windforce) and after reading the numerous post/guides on here I have reached what I believe to be a stable overclock.

But I was wondering how long I should run the stress test, and what test should I do/use before I declare it to be 100% stable?

I am using EVGA Precision X as my oc utility and EVGA OC Scanner X as my stress test utility


I must be extremely lucky or its just not stable enough yet for gaming because my OC looks like the following so far


107% Power Target
80C Temperature Target
+135 mhz - GPU Offset
+350 mhz - Memory Offset
1200 Mv (stock)

 

[LightningJR]

100% is nearly impossible for overclocks since there are many variables. To get a good overall feeling of stability I would not just use a stress tester but also play the games you normally play. A dedicated stress tester is not a good representation of how games use your GPU and will not give you the best answer to how stable an overclock is.

I have had Furmark going for 2 hrs and then Heaven benchmark fail in 2mins and other games fail under 5 min.

Do a mix of everything and if it doesn't fail, even on the warm days, you should be doing ok.

 

[FireFox]

Overclocking, Power Draw, Noise and Temperatures

Overclocking the EVGA GTX 770 is just as easy as overclocking the GTX 700 and 600 series using PrecisionX. What is surprising is that we were able to overclock we. past our reference maximum base overclock of their GTX 770 SC at 1111/7010MHz to 1176/8210MHz. We regularly saw boost to 1306MHz although temperatures never quite reached 80C – even at maximum load.

We were able to add an additional +65MHz over the EVGA base clock for a total offset of +135MHz over the 1046MHz base clock that Nvidia set for the reference GTX 770. On top of that, we were able to get +300MHz on the memory. We were able to go higher on the memory but stopped when we were no longer getting performance increases.

We did not adjust the EVGA GTX 770 SC’s voltage. Temperatures were never an issue and the fan profile remained at stock which meant that the GTX 770 SC is also extraordinarily quiet at maximum load – the fan profile rarely went to 60% and the temperatures remained well under 80C under the most stressful conditions and highest load in any game that we tested. The EVGA GTX 770 SC and the reference GTX 770 are both very quiet cards but the ACX cooler is quieter, especially under load.

Let’s head to the performance charts to see how the EVGA GTX 770 SC compares with the reference GTX 770 each at stock, Superclocked from EVGA’s factory and overclocked, as well as against the stock (925/1375MHz) and GHz Edition (1050/1500MHz) AMD HD 7970. As an added bonus, we are going to overclock the EVGA GTX 770 SC as far as it can go without raising the voltage or adjusting the fan profile to see if we can narrow the performance gap with an EVGA GTX 780 at reference speeds.

This is a Guide that i found and used months ago to Overclock my 770.

 

[alwayssts]

Maybe I'm weird, but these days I mostly just run Heaven (or heaven plus one of the msi tests...I never run furmark etc alone because some of them are purposely throttled by nvidia's drivers and if using them by themselves it can lead to false positives). I don't know if this is anyone elses SOP, but I look for the following once gpu has leveled out in temp:

1. Artifacts by the canon when the camera moves through the ship. (blue/green lines? I forget exactly.)
2. Artifacts (black specs) when it starts the overhead 360 of the statue (usually visible by or out to the right of the wings).
3. Rendering oddities on the flag/canon after the going through the courtyard. The scene on the top of the castle is also a good way to test load throttling (it's pretty intensive).

It's not a catch-all, especially for memory (which has become a weird thing because of error correction...basically the best way to test that is by upping it until benchmarks go down afaik).

Lightning is right; nothing is a catch all because different games stress different aspects of the gpu at different times, and certainly ambient temp can matter. Nothing is really guaranteed to work all the time (over stock). That's usually where I start, though, and for core clock it's worked pretty consistently. For mem I usually dial it in with a less precise, rounded-down manner after seeing what will/won't cause any massive oddities/crashes over a couple hours of gaming, and then back it up with couple bench runs while comparing scores. Typically though, given the problems memory *can* give, I error on the conservative side in that respect. It just isn't work the hassle as the performance gain over bandwidth needed is generally minimal, and most cards are not limited by bw to the point you need every last mhz.

 

[acvolta]

Given that my card (Gigabyte Windforce) is/was factory OC'ed I was not able to put much of an overclock on the card as a result I hit a max "stable" OC in Heaven of the following

110% Power Target
80C Temperature Target
+80 mhz - GPU Offset
+300 mhz - Memory Offset
+12 Mv - Voltage

 

[DarkOCean]

Give it to me, I'll stress it out for 2-3 years just to make sure and then give it back to you.

 

[CAPSLOCKSTUCK]

Id rather test my builds by using them rather than stressing the bits in them.

If ever i run somevthing like p95 or Heaven ill monitor temps with hwmonitor.

I love reading shit about overclocking adventures though.

I cant remember the last time i did something on a pc that was so important and cpu intensive that a cradh would worry me. Different if your cranked up to the max on everything and doing work.

 

[acvolta]

Id rather test my builds by using them rather than stressing the bits in them.

If ever i run somevthing like p95 or Heaven ill monitor temps with hwmonitor.

I love reading shit about overclocking adventures though.

I cant remember the last time i did something on a pc that was so important and cpu intensive that a cradh would worry me. Different if your cranked up to the max on everything and doing work.

While I tend to agree with you, testing them your self in ones own work/gaming environment it takes allot more trial and error to get the system working...not to mention that it my not be stable with every game out there so I stress test just to ensure 100% compatibility and also to save time from messing about with it.

 

[xLegendary]

Furmark is usualy pretty heavy on the GPU load.. you might try a couple of hours

 

[natr0n]

Play a few rounds of BF4 if you have it. That game is extremely sensitive to unstable overclocks.

 

[jaggerwild]

Yeah,
Use furmark to burn it out, put undue stress on it great suggestion!!!!!

 

[Steevo]

Yeah,
Use furmark to burn it out, put undue stress on it great suggestion!!!!!

Yeah, cause modern cards don't have thermal or power limiters built in to prevent that, great job and most useful comment here!!!



Furmark is a good tool to check for unstable overclocks, we have used it for years. Cinebench is also good as it checks the rendering for errors, but GDDR5 will or should not show errors beyond the missing texture when the data is corrupted in transfer. You will however either see the missing textures, a drop in FPS as the data is retransmitted, or the striped screen or grey screen of death when the card hard locks.

 

[LightningJR]

Yeah, cause modern cards don't have thermal or power limiters built in to prevent that, great job and most useful comment here!!!



Furmark is a good tool to check for unstable overclocks, we have used it for years. Cinebench is also good as it checks the rendering for errors, but GDDR5 will or should not show errors beyond the missing texture when the data is corrupted in transfer. You will however either see the missing textures, a drop in FPS as the data is retransmitted, or the striped screen or grey screen of death when the card hard locks.

Ahh, I never thought of using Cinebench for that, great suggestion. I'll remember that, thanks.

 

[FireFox]

Furmark is usualy pretty heavy on the GPU load.. you might try a couple of hours

Something that I've learned from a friend of mine who has a Pc shop (best way to stress your Card is Playing Games)
You stress your GPU with Furmark or some other GPU stress tool per hours and everything works fine but could be that after a couple of minutes playing Games you could get errors.

Yeah,

Use furmark to burn it out, put undue stress on it great suggestion!!!!!

You might also suggest to fry it.

 

[CAPSLOCKSTUCK]

I have managed to feck up GPUs and CPUs and have definitely shortened the lifespan of others. Smoke is generally a pretty good sign that "oops i shouldnt have done that"

There is thermal throttling and other built in safety stuff but you can still wreck your stuff.

Stress test .....yes
Monitor carefully .......yes

Stick on P 95 and piss off for the weekend........... no

 

[The N]

Furmark and p95 both worked well for GPU stability but too much unrealistic load, which suggest stability in extreme conditions. on the other hand. i do think few benchmarks like Valley 1.0v, Heaven 4.0v and in game benchmarks are Good too. shorter and good stability tester for overclocking.

 

[Blue-Knight]

How long should I stress test my GTX770 to ensure its 100% stable

Until its temperature do not change much in 5-10 minutes and you do not see artifacts.

Do that running furmark.

Good luck!

 

[GorbazTheDragon]

For GPUs I just wang up the volts and clocks, give it a quick check in furmark, so I'm sure it will stay within a reasonable temperature range, then start gaming. If it crashes I slowly turn down the clocks until it is stable. If your GPU is unstable it is (usually) not nearly as bad as when your CPU/memory is unstable, as it doesn't give BSODs most of the time.