Weird overclocking behavior

I can't wrap my ahead around this. I used Unigine benchmarks to find stable clocks, which were +125 on core and +275 on memory. Here's the thing, though: the core overclock is unstable to more or less degree in every game. To make things even more complicated, how the instability behaves is just plain weird. For example, if I use +125 and start RE2, GPU load shortly spikes to 100% and game crashes during startup. If I lower it to +75, game starts and runs just fine. That being said, if I increase it back +125 while the game is already running, it's all fine and dandy. However, if I try to restart it with +125, it crashes at startup again.

The OC stability varies from game to game, hell, even from day to day. Evolve, of all games, were running fine with +75, but yesterday, out of the blue, crashed even with that OC. GPU simply spiked to 100% and bam - CTD. SotTR is allergic to anything above +75, etc.

I have never experienced such inconsistent OC behavior in my life. Were I just lucky all those years and this is the normal thing?

 

quite common, means the power doesn't transition fast enough to serve the clock speeds.

 

Is that a PSU issue?

 

+125 is not stable, end of story.
i can also overclock my card and run many games without a problem, but there is a game that shows glitches with that setting, so i have to downclock a bit.

i found a overclock that runs all settings flawlessly. call it rockstable or something

 

Overclocking is always snakeoil on modern GPUs with already High-Clocks.
5% more performance is basically a Placebo Effect.

 

Hi there

I don't personally use Unigine Superposition or Heaven or Valley there, these benchmarks are not most reliable benchmarks for OC there

In these Unigine benchmarks I remember I could clock my GTX1080Ti to 2164MHz quite easily but in games I couldn't run such clocks, similarly I couldn't sustain or run such clocks in games, usually I run 2113-2139MHz in gaming, in some games I run 2113MHz at 1.07v

Great for OC is using something like is FireStrike or Timespy etc or Octane Benchmark in which I usually spot quite easily instability or rather play games and try different OC

OC can vary from game to game and sometimes this depends too on Silicone lottery of your GPU as well

On my RTX 2080Ti usually I run for non RT games or games which don't have RT I'm running 2130-2145MHz OC(SOTTR or ROTTR, Witcher 3 and many other games) on other games which does support RT then OC can vary but usually I'm running 2085-2100MHz in these games, in RT games like Control, Metro Exodus, BFV etc I simply can't run higher clocks than 2115MHz tried everything, but usually I'm running 2085-2100MHz which are stable in these RT games and no issues with instability

Best benchmark is actual game or rather some benchmark which I mentioned bit everyone have own benchmark in which he can spot instability

I usually use for spot instability for GPUs rendering which does support OptiX like Octane, Blender Cycles, IRAY etc and then I try actual game if OC will or is stable and then 3DMARK FireStrike and Timespy

Hope this helps

Thanks, Jura

 

no, its a board delivery issue.

 

I guess it depends how far headroom has already been pushed by the manufacturer itself.
I know some graphics cards are already clocked very close to limits straight from launch. These cards are usually noisier and have high temps, usually come at the end of the series or as rebrands.
Direct example of clocks being changed closer to limits probably is stealth bios spec update to those AMD cards recently. Even thou those cards are still in acceptable noise/temp range.

 

I think it's based on how rtx does turbo boost or whatever they call it . It is dependent on temperature so when you start the game it might be boosting +125 over what it thinks it's stable at that temperature so let's say it's 2000 + 125 at that temp =2125 but in reality 2125 is unstable . As the temp goes up let's say it boosts only to like 1900 on it's own as a base and then it ads 125 so 2025 is actually stable. You can edit the curve in MSI Afterburner but i would suggest just using the auto scanner.

 

Given that have not told us what your overclocking we`ll have to assume its your MSI 2070S GAMING X trying to overclock, you also have not mentioned temperature of your gpu, I have the 2070s Ventus OC which has a lower boost clock than yours so i`d say your past your limit.

 

its about clocks spikes...
thats behavior boost 4.0
in heavy scenario you have 2000MHz but in light work scsnario you have 21250and GK its not manage this clock
my 2080ti managed 2150MHz but problem is this spikes, and result is usable only 2075Mhz
honestly its not worthy hihger PL and voltage for few %
all this cards about balanced between temp + clocks + PL

 

I just use MSI Afterburner, presst Crtl+F, then I just let it do a Scan OC, a good, stable boost for any game and just set my memory clock to 190-200+. That's the most extra performance you will get out of Pascal and Turing, if you want higher fps and stable games, you just gotta get a stronger GPU.

 

Many people are unaware that their overclocks are actually unstable until one of the games crashes. The benchmarks or games they were using didn't really stress their high end GPU at all. For memory overclocks I use Kombuster with artifact scanner. But when it comes to core clocks you just need to play demanding games made with different engines. Hitman 2 was the one game that broke my "stable" overclock. If your CPU isn't good enough try increasing the resolution scale until you're GPU bound.

 

just play TES:4 Oblivion, if the gpu's unstable everything will turn green

 

With metro 2033 and LL if gpu is unstable everything turns red.

 

Whoa! I didn't notice all these replies. Thanks, guys.

 

for the Turing cards it's better to look at the Frequency Curve graph.. not just the clock offset. Under very heavier load and power draw the card might not be able to boost as high so it will run a lower clock speed for example in RDR2 if I'm at over ~85% GPU usage my card is basically sitting at or near the power limit and starts dropping my clock speeds somewhere below 2000Mhz. Under these situations it's perfectly stable as the clock and voltages drop down to stable settings. The issue with higher offsets typically crop up when your GPU load isn't as high and the card can boost to it's maximum clock speed on the frequency curve. So if you encounter an area in a game that isn't that demanding to render and say GPU load drops to 50-60% or lower and suddenly the card boosts up to the max voltage of 1.15v and whatever maximum clock speed the curve allows for. If that maximum clock isn't stable you'll get a crash.

Now when the game is loading or if your in menus that don't heavily load the GPU the clock speed can ramp up to the highest point on the frequency curve.. which if your offset takes it too high can result in a crash. Of course some games hardly load the GPU at all in menus and the card will drop to the base clockspeed. Just depends how demanding the menu is regarding a 3D graphics load. My card is already factory overclocked +120Mhz from MSI over the stock 2080 Ti. So my card can only manage an additional +25Mhz as that puts it at 2085Mhz maximum clock at 1.15v. You can see below on the frequency curve editor.



If I set a very high offset +60Mhz which in reality is actually +180Mhz overlock it puts my card at 2130Mhz @ 1.15v maximum clocks... Which will result in a crash. Personally I've found my card just gets unstable at any clock speed above 2100Mhz. So I set my curve so it never gets above 2085Mhz.

 

I never got success with freq/volt curve (always unstable at some point in games so frustrating) and with auto scan (giving me less OC). Only my manual OC was a success and stable at high freqs in games.

 

The best thing you can do with current gen GPUs is to undervolt them. My card would do ~2050 max out of the box, I then limited this to ~1950 but with a significantly lower power usage.
One advantage is the stable performance, as the card runs at one speed under load, at all times, the other is significantly less heat and noise.
I'm running a simple curve, which is flattened out above my voltage limit and slightly raised above regular below. It's all about efficiency

I'd only try to run max speed at max volt with water cooling or noise protection. The difference in performance is negligible anyhow nowadays.

 

Hi there

Most of Turing or Pascal GPU are limited by 1.094v max voltage, there are some BIOS which allows you to run 1.15v like Asus XOC BIOS for GTX1080 and Ti, Kingpin or HoF Galax 2000W BIOS or NDA BIOS from MSI

On normal GTX1080Ti or RTX 2080Ti or RTX 2080 etc you are limited by 1.093v to 1.094v voltage limits, due this is no point manually creating V/F Curve in MSI Afterburner beyond 1.093-1.094v

Hope this helps

Thanks, Jura