[SOLVED] How to tell if my GPU overclock is unstable

[BiscuitSole176]

Hello! This is my first time ever overclocking a graphics card. I have a RTX 2070. I used the automatic scanner from the MSI app to achieve around about 190 MHz on core speed and increase my memory to around 580 MHz. I am paranoid as far as causing damage and wanted to get input on how to tell when too much is, too much. I have played a few matches of War Thunder completely maxed out with nothing out of the ordinary occurring. I also ran the MSI Kombuster and saw nothing that stood out, my GPU temperatures stayed between 65-70 degrees Celsius.

 

[Newtonius]

Overclocking can always be scary at first but rest-assure that you cannot damage your card unless you over-volt your GPU which at this point msi afterburner (the app I assume you're using) has disabled this feature. You can use GPU stress testing applications like Unigine's Siperposition application to see if your GPU runs stable during full load tests. If the application crashes or you get any errors then that's how you know you've over-exceeded your cards limits.

If you do get any application crashes after overclocking don't be worried as the card will not suffer any damage. Just reset the core and memory clocks to default and try again. You may also raise the overall power-draw (the second top option in MSI afterburner, this ups the WATTAGE, not the VOLTAGE) to 110% as the card's VBIOS (your GPU's firmware) will know it's safety power limits and pull more power but to a safe level.

Also note that the better you keep your card cool (64c or below) the best and more stable performance OC you can achieve. So keep an eye on how far you can push your card depending on your avg GPU temps during gameplay.

tl;dr
Overclocking doesn't essentially damage the card, only changing the VOLTAGE will cause long-term affects.

 

[drea.drechsler]

Hello! This is my first time ever overclocking a graphics card. I have a RTX 2070. I used the automatic scanner from the MSI app to achieve around about 190 MHz on core speed and increase my memory to around 580 MHz. I am paranoid as far as causing damage and wanted to get input on how to tell when too much is, too much. I have played a few matches of War Thunder completely maxed out with nothing out of the ordinary occurring. I also ran the MSI Kombuster and saw nothing that stood out, my GPU temperatures stayed between 65-70 degrees Celsius.

The one thing I've found is 'stress tests' for GPU's don't work. Superposition, Valley, Heaven not even 3DMark stress test make my overclocked GPU bug out as well as Folding@Home does.

Yep...just go get Folding@Home, easy download, easy install. Start the folding process and let it run a while, then go check the logs in the Advanced Control applet. A bit geeky looking through the log, but if the GPU is overclocked too far it will start throwing out bad results and it will show up as errors. The calculations show up bad results faster and easier than waiting for a crash in a 3D application.

 

[Phaaze88]

Yep...just go get Folding@Home, easy download, easy install. Start the folding process and let it run a while, then go check the logs in the Advanced Control applet. A bit geeky looking through the log, but if the GPU is overclocked too far it will start throwing out bad results and it will show up as errors. The calculations show up bad results faster and easier than waiting for a crash in a 3D application.

I may have to try this out myself. About how long would you say is enough for stability?

 

[drea.drechsler]

I may have to try this out myself. About how long would you say is enough for stability?

It's very critical about results accuracy, for me it ran only about 15 min's for it to start rejecting results. After giving it some more voltage and backing off on the memory clock I let it run for over 4 hours, turning in three or 4 work units complete, and it was perfectly good.

It doesn't show you it's throwing errors, except if you watch the completion bar in the web control window it will reset occasionally when it rejects a unit that has errors. F@H is much better than some of the other BOINC projects about telling you your system is putting out errors at least. I worked with one that would take the results, hold them until two other computers did the same work unit and compare results. You'd get sometimes weeks of work piled up before they'd finally come back and tell you it was bad.