Why does RTX Enabled crash when GPU is overclocked?

[FinalLightNL]

Hi there,

I have a RTX 2070 Super that is normally overclocked to +100 core +1200 memory.
It is stable, never had any problems with it at any game for months.

But with cyberpunk it simply crashes my game when I enable ray tracing, disabling it is just fine zero crashes.
Same with stock settings of my GPU, enable RTX no crashes.

It just crashes even with the slightest OC.
What might cause this?

EDIT: 1440P everything maxed and ray tracing all on & medium lighting i get around 50~55fps without my OC.
DLSS at balanced.

But when i enable my OC i get over 60fps constantly, but the crashes are just frustrating

 

[TheMansitos]

Same here: RTX 2070 Gaming Z with +60 and +600, used this OC for 2 years with actually no crashes in different games.
Btw if my gpu is overclocked in cyberpunk (also tried lower oc profiles) the game keeps crashing. I think that RED engine is very sensitive to OC, I remember having the same problem with my GTX 970 in The witcher 3.
It's sad because with my OC i get around +10% fps... but I need to play with stock clocks in order to eliminate crashes...
Maybe it's not the engine but actually the OC is stable in other titles because they are not using RT cores. As soon as you use them, the load is very high for the gpu and the oc become unstable?

 

[plankdalf]

there fore its not stable.......
how many actually benchmarks have you done and what ones?
because if its just 3dmark then... sorry mate deffo not stable, 3dmark is so optimised that even the most absurd and highest oc do not crash the runs.
i have a 2080ti and i clocked 2500 mem and 260 plus on core (bare in mind it water cooled hence why its so stupidly high) in any other games it was fine including benchmarks and yes fps boost was ok till it wasn't

keep a eye out for points loss in benchmarks do the same test over at least 10 times with your lowest oc and just jot down the score numbers and estime yourself a avg for it. then test your highest and if its consecutive scores are lower..... hey ho your OC is deffo not stable and its by far the most easiest way to note it.

so as soon as you start loosing points in the benchmarks i suggest dialling it down by 10mhz separately on core and memory

 

[Azulath]

As @plankdalf has mentioned, Cyberpunk 2077 crashes because your OC is not stable, simple as that. Also, enabling RT also puts strain on your RT cores which normal rasterisation does not. Hence, more strain is put onto the GPU which may result in crashes.

 

[andrewdilley]

Further small side comment to the OP regarding OC stability: I had same as you after an earlier patch (forget which one, before 1.3 I think) and blamed CP2077 for being broken, since I had no issues on other games and suddenly it was repeatedly crashing after a few minutes. But then I dialled-down the modest OC (I lowered the Power Limit) and everything ran fine.

BUT - in 1.5 I've found I can go back to my original OC if I want, with no crashes (so far, touch wood).

Conclusion: I guess any version of any game might unexpectedly break a previously 'stable' OC.

 

[Azulath]

Any OC is inherently unstable and may lead to issue. Running a stable OC is extremely hard and it might be that a card that is able to run at certain settings might not be able to do that a year later. Thus, one should always be able to adjust an OC after some time has passed and also on a per-game basis.

 

[Bealdor]

The most taxing benchmark for OC stability I've seen till now is the Metro Exodus one in a 10x loop.

If your OC survives this, it'll survive every game out there.

 

[Bassdeff]

I don't know what it is about CP2077 and Metro Exodus but those are the only 2 games in my library that give me issues with my Gygabyte RTX3080TI Gaming OC (left at factory settings) and the only thing that fixes it is down clocking the GPU by 100mhz. When I do that those games run flawlessly. I ran a number of different torture tests and benchmarks (3D Mark, Unigine Heaven and Superposition, Cinebench and Prime95) when I built my system and I've never had any issues with any other games I own but there is something about those 2 games that GPUs don't like. I had the same problem with my Strix 1080TI OC (left at factory settings) on my previous build. I lose a bit of FPS by down clocking but I'll gladly give up 5 or 6 FPS to not deal with crashes.

EDIT: I discovered a little tid bit while snooping around some forums. If you have "prefer maximum performance" enabled in the nividia control panel not only does it lock your idle speed at base clock values but it also throws the boost clock way up. With it turned on I was peaking around 1985mhz (usually where the crashes happen) before dropping back down to mid 1800s. I turned that setting back to normal and ran an OC Scan with afterburner to get another curve and did some more tests. The game runs stable with clock peaking at around 1890mhz and remains in the mid 1800s fairly consistently and my GPU seems to be using a little less power. My curiosity has been peak and now I'm going to see how high I can push it with power management set to normal. I never really paid attention to it. For years the first thing I would do when setting a game profile was set that to max performance

 

[Notserious80]

Yea i tend too leave power settings at normal on my 3090 too. Just seems too crash less with it on for some reason. Tested out the oc scanner on afterburner now too and hade a crash ofc ^^ but second time i got it too 100% and the oc was really small but still maxed out my power limit... and i scored higher then with my normal oc even tho i was like 100 mhz lower on the oc. Something weird is going on....

Edit: Nvm. Reseted the curve and now its even better with my oc. Something must have been messing with the Boost and power limits.

Edit2: Tested some more. In syntetics i can get a +175 core with +1500 mem oc pretty easily. In cp 2077 anything over +100 core just crashes. Atleast its better then at release when just touching anything crashed the game. Not sure why it crashes either since temps are fine and im generaly not close power limit. Guess it can spike possibly but o well.

 

[Bassdeff]

@Notserious80

Unfortunately for me increasing the max clock and memory any higher than whatever gigabyte did on this card is very very neglible due to the 350watt power limit (375 watt peak) for my card. Which is why I've been running it at factory settings although I've been playing around with voltage curve that afterburner has given me and I've been getting some interesting results. It massively undervolts the lower end of the frequency curve gradually getting closer to the factory voltage curve as clocks increase. I get a small 4fps boost in with the superposition 4k benchmark but it is a much more stable FPS than I had before. Still have to extensively test it in game and I might have to tweak it some more but so far it looks pretty promising as I've been seeing fewer big spikes over the power limit. ​

​

 

[Notserious80]

Yea thats kinda a hard limit that you cant exactly get past without shunt modding or hacking bios possibly. But if its a 2x8 pci-e plug card even that wont do anything since you cant get more power too it. Kinda does very little too OC these cards anyways when it comes too gaming. I got 1500 higher score with my oc in superposition at 175/1500 OC witch is not nothing but ingame its like 4-5 fps. Also the game gets more unstable.

 

[Azulath]

I have also undervolted and a tiny bit overclocked my 3090 with the following curve:

Setting the clock to 1920MHz @862mV leads to crashes in Cyberpunk 2077, but runs fine in any other game I have tested.

With that being said, I would neither touch the power limit, due to the New World debacle even though I do not have an EVGA card, nor the memory clock, due to the high memory junction temperatures, with a ten foot pole.

 

[Notserious80]

Yea sadly the 3090s back side memory was a big oversight from nvidia... But im not sure that was what really happend with those cards. I solved the heat issue so i dont worry about it any more but with original cooling it gets very hot. Also you can monitor those temps with Hwinfo but any OCing has its risks. They also should have downclocked at 105c if the card worked as intended.

I never se anything above 60c now with my waterblocks so im able too OC pretty hard and hit power limits without getting close too temperature limits so im not worried but if its woth it? not so sure ^^

 

[Azulath]

Yea sadly the 3090s back side memory was a big oversight from nvidia... But im not sure that was what really happend with those cards. I solved the heat issue so i dont worry about it any more but with original cooling it gets very hot. Also you can monitor those temps with Hwinfo but any OCing has its risks. They also should have downclocked at 105c if the card worked as intended.

I never se anything above 60c now with my waterblocks so im able too OC pretty hard and hit power limits without getting close too temperature limits so im not worried but if its woth it? not so sure ^^

Ah ok, a waterblock definitely solves the issue I plan to go full custom loop on my next build (RTX 4090 or 7900XT) but this one only has an AIO for the CPU. Luckily I mounted my GPU vertically and I have three bottom case fans which blow cool air right through the GPU. Furthermore, I have attached heatsink blocks to the backplate which also decreases the temperatures a bit. Without those blocks, I was hitting 86°C (sometimes 88°C) and now I'm sitting at more comfortable 82-84°C. Not much, but beggars can't be choosers XD.

While the maximum temperature of those memory modules supposedly is 110°C (or something like that), Micron and Nvidia have been a bit sketchy in terms of information disclosure for my taste and I have also heard of people whose memory modules died from constantly running at only 95°C. Thus, I prefer to keep mine as cool as possible, but I won't be using a waterblock for this one.

Anyway, I think if we want to discuss this further we should open a separte thread, otherwise we will be derailing this one.

 

[Bozobub]

Note that *external* overclocking isn't very comparable to *built-in* overclocking options. For example, my mobile 2060 RTX is fully stable on max CPU and GPU o/c, enabled either by engaging the included "Turbo" button or setting max overclocks in "PredatorSense" (the Acer Helios 300 management software), but setting the same clocks in an external program easily crashes ^^' .

As noted above, your o/c really isn't all that stable, just *stable enough* for lighter loads.

Re: increasing TDP/power limits: AS LONG AS your temps are fine, this won't easily harm your system, but this also can greatly compund any damage if you accidentally run too hot, which varies quite a bit per individual GPU/memory combo. Overclocking is NEVER without risk, even using built-in options (not counting CPU "Turbo boost").

 

[Azulath]

Note that *external* overclocking isn't very comparable to *built-in* overclocking options. For example, my mobile 2060 RTX is fully stable on max CPU and GPU o/c, enabled either by engaging the included "Turbo" button or setting max overclocks in "PredatorSense" (the Acer Helios 300 management software), but setting the same clocks in an external program easily crashes ^^' .

This is because there can be different interpretations of an overclock depending on the system state. If the button is changed/clicked, the OC is applied at the boot with probably a different BIOS being used. Doing an overclock through software results in the stock settings being shifted and they might be shifted differently.

As noted above, your o/c really isn't all that stable, just *stable enough* for lighter loads.

Re: increasing TDP/power limits: AS LONG AS your temps are fine, this won't easily harm your system, but this also can greatly compund any damage if you accidentally run too hot, which varies quite a bit per individual GPU/memory combo. Overclocking is NEVER without risk, even using built-in options (not counting CPU "Turbo boost").

Yeah, but there can still be issues and I would not touch the power slider on a 3090 (unless you know what you are doing). Most other cards should be fine, but as I've already mentioned, Ampere does has issues with transient load spikes.