Overclocking GPU without overvolting - risk free?

[Alpha0mega]

I was considering OCing my GTX 970 setup. I have a couple of Zotac 970 (vanilla) cards. Doubt they will be great overclockers, but would still like to find out if I can get better performance out of them.

Since I have never OCed GPUs before, I am a little hesitant. During my research into the possibility of harming the cards, I ran into something interesting that I would like the community here to confirm/refute.

It’s the PC Gamer interview/tutorial with an Nvidia engineer, Tom Petersen, regarding overclocking SLIed GPUs.

At the 13:40 mark (linked below), the interviewer starts talking about upping the voltage (they were sticking to non-voltage adjusted oc’ing up till that moment), and Mr. Petersen says “voltage is when we start transitioning from free and you can’t hurt it, to there is a cost. So if are playing with any of these slider like the… the GPU clock or the memory clock or the temperature, all that stuff is 100% guaranteed to not hurt your GPU”

https://www.youtube.com/watch?feature=player_detailpage&v=TF_89emuI9k#t=819

I was a little surprised by this. I know that voltage based OC’ing is more risky compared to just increasing the core/memory clock, but is the latter really “100% guaranteed to not hurt your GPU”? Is that really the case? What happens if one goes overboard with the increments to core and memory clock? Just temporary visual corruption and/or stability issues, which disappear when the values are brought back to sane level, without having any possibility of hardware damage?

 

[Silverforce11]

Modern GPUs have built in safety, when pushed they will hard reset, either a driver crash & reset or a system crash requiring a reboot.

If you don't touch vcore, its very safe.

 

[n0x1ous]

and even if you do touch vcore, keep it under 100mv and its still pretty safe!

 

[ShintaiDK]

Overclocking is never safe.

However overvolting is the biggest danger.

 

[.vodka]

Well, there's no harm in going for a stock volts OC on your cards. You'll get extra performance with almost no added heat or fan noise.

Once you start overvolting, diminishing returns quickly kick in. Your 970s are cool running cards and their coolers can probably handle the extra heat. A few mV aren't a problem (+25, 50 mV for example) if you can keep temperatures under control, heat is the main offender here. Now, if you go nuts on the voltage without the cooling required for it (not only the GPU but also the VRMs) then start worrying about reduced lifetime for your card and/or the risk of it breaking down. Not worth it IMO.


It's all about finding the sweet spot for your hardware. A little over stock volts are sometimes all you need. Once you need to add ridiculous amounts of voltage for some extra MHz then it's not worth it anymore. As silverforce said, there are lots of safeguards in modern cards, it's quite difficult to kill one of them by only overclocking. Download your favourite OC utility (Afterburner is a good one) and some program to test (unigine valley for example) and see what your cards can do before artifacting and/or locking up (clear signs of instability).



In my personal case, my R9 290 Tri-X comes with a 25 mV factory overvolt, 1000 MHz stock core frequency (up from 947 mhz on the reference 290). With that voltage it can do 1075 MHz and runs almost as cool as when at stock clocks. With +50 mV it can do 1100 MHz, maybe some more but you can now see the frequency to voltage relationship is not as good anymore... as expected of Hawaii, at least.

Before that I had a XFX 6850 that ran at 775 MHz and did 920 MHz on stock volts, 1.25v (DMM measured). On a side note the 6870 ran at 900 Mhz so nothing out of the ordinary there. I was curious about what it would need for 1000 MHz, so I did a little hard mod... about 1.35v. Not worth it at all. Anyway, I had that card for almost 4 years and ran it 24/7 at 920/1180 from 775/1000. It's still working just fine in a secondary rig.



Be reasonable with your overclocking and overvolting, you are pretty guaranteed to never kill your hardware.

 

[Alpha0mega]

Thanks. I am going to go for it. Going to overclock without overvolting. Tom Pretesen is an engineer for Nvidia, so I am going to assume that he is correct, despite it being an argument from authority.

 

[KaRLiToS]

As long as you keep the temperature down and safe voltage. With stock BIOS, most of the time you are safe to crank the voltage to the max.

Most of the cards have locked voltage, giving a safety. What I usually do with my cards to check their potential is turning the power tune to the max, voltage to the max ( with stock BIOS that is ) and raise the fan speed and try to reach highest clocks. Never encountered any issues yet.

I have personally cranked 1.6v in all my R9 290x for benchmarking purpose for quite long period and I always kept the temperature below 55'C. Never saw any sign of degradation. ( that was with BIOS flash)

 

[Ed1]

Overclocking is never safe.

However overvolting is the biggest danger.

So what about when manufacture over volts ?

I see you have 980 ,some 9xx run 1.175-1.20v or even lower but some run 1.22 or even more (MSI gaming, G1 are two ).

 

[ShintaiDK]

So what about when manufacture over volts ?

I see you have 980 ,some 9xx run 1.175-1.20v or even lower but some run 1.22 or even more (MSI gaming, G1 are two ).

Its still not safe. It just comes with a warranty however.

 

[Face2Face]

As long as you don't install a custom BIOS and start throwing massive amounts of mV at it; it's fine.

Modern day NVIDIA and AMD graphics cards have built in protection through the use of Dynamic clocks speeds, voltages, and the use of throttling. If your GPU gets too hot; it throttles. Pulling to much power; it throttles. For no apparent reason at all; it throttles. Overclocking nowadays is easier and is much safer than it ever was.

 

[Deders]

Be careful when overclocking the memory, this will be the most easily damaged. Overvolting the GPU will most likely cause damage over a long period of time (if at all) where you won't be able to reach the same high clock speed you thought was stable.

 

[Alpha0mega]

But will overclocking the memory, without any overvolting, still possibly damage it? That's the crux of my problem. The Nvidia engineer is stating that as long as the voltages aren't increased, there is absolutely no chance of damage.

 

[.vodka]

I don't see how the memory can get damaged only by overclocking. Not in this day and age.

If you push it too far then you'll get artifacts and/or a black screen, in that extreme case you just need to hit the reset button, and there you go, you have the upper limit for your particular card's memory. Then it's not hard to find the highest stable frequency with that in mind.

 

[thxdd]

I have never seen any core or memory damage done to a GPU unless bumping voltages or intentionally running above thermal limits for extended periods of time (almost impossible to do with throttling nowadays). If you aren't overvolting or flashing custom BIOS files, overclocking nowadays is virtually risk free.

I recommend bumping up in small increments (core first, then memory) and benchmarking/load testing thoroughly with each increment to validate stability. I usually start first with overclocking the core, then move on to memory.

 

[Silverforce11]

These days even if you crank vcore it will hard crash and reset if its beyond what it can handle. As @Face2Face mentioned, both sides have been pretty good about this.

A few generations ago it wasn't the case, if you add vcore even with the default bios limit, your VRMs could explode (gtx570 and 590!).

GDDR5 will artifact or crash when pushed.

 

[Alpha0mega]

Well, I finally went for it and OCed my cards. They are now running at core +225 over stock, and memory +325. Tested with 2 hours of Unigine Valley at max, and 1 hour of Firestrike. In Valley, I could push the memory to +350 and beyond without any artifacts, but Firestrike showed an occasional green flicker, so I brought it down. All this is without any additional voltage at all. I am actually quite pleasantly surprised, since I didn't expect the cards to OC so well.

I am getting 10% higher scores in 3DMark Firestrike and Skydiver, but only 5-6% higher in Valley and Heaven. Is that normal? It's not because of throttling, since I am monitoring the core and memory clocks, and they remain high, with normal variation, throughout the run.

 

[FX2000]

When you overclock something, it either works 100%, runs with stability issues, or just does not work. Putting it back at stock will always fix that. When you add them volts, it either works & draws more power(heat), or you done f**ked up and gave it too much volt goodness, and the magic smoke will escape the electronics, removing the magic fairy smoke they run on.

 

[Deders]

Ram is more sensitive, If you leave it running tor too long at higher clocks than it can handle it can stop working all together, much quicker than a GPU would. I also find that some games will cause artifacts at lower clockspeeds than others.