Question EVGA RTX 3060 XC 12GB used gpu FREE gift very poor overclocking performance

[realflow100]

I got an EVGA RTX 3060 XC 12GB as a FREE gift from a friend. who got it from another friend that used it for crypto currency mining. and when I tried to overclock it I get very wimpy increase in offsets.

+100% voltageslider
+11% power limit slider

+140mhz core offset
and only 300mhz memory offset.

I get artifacts very rarely at +400mhz memory offset. every once in a while. about 1-2 minutes I'll see a stray polygon flash on the screen for a split second. Or a stripe. or rectangle. or blue colored polygon for a single frame.

However if I keep going up. It doesnt rapidly increase even up to +900mhz. its very gradual and its only up to 1 artifact every couple of seconds (obvious enough to see quickly but not crash the GPU)

FPS continues increasing up to +900mhz without regressing or stuttering. just more noticable artifacting every few seconds.

Core clock is unstable beyond +140mhz. but I think I may have to decrease it another +10mhz if it crashes again. it crashed after about 10 minutes in heaven benchmark at +150mhz

The temps all seem fine. hotspot isn't too unusually high.
GPU Core maxes around 69-71*C under full gaming load. Even with the fans maxed at 100% fan speed and rpm
At idle the fans spin down to 0rpm with my custom curve and my gpu reaches about 22-26*C when sitting idle for 20 minutes.

How are people getting even beyond +1000mhz memory clock offset? Mine doesnt even get high enough to the point it starts error correcting and dropping frames.

Should I maybe replace all the thermal pads and paste? Maybe the old owner messed it up or something when cleaning it? or applied the paste poorly?
maybe the old pads are dried out and cracking?

 

[Unolocogringo]

It could be any or all of those things.
The card was mined on which usually pushes memory Speeds/ temps to the limit. For long periods of time which degrades the chips.
Many mining cards will not even run default speeds any more so be glad yous will overclock at all.

 

[Phaaze88]

How are people getting even beyond +1000mhz memory clock offset? Mine doesnt even get high enough to the point it starts error correcting and dropping frames.

1)They don't really know what they're doing. Some people 'copy-paste another person's answer sheet and don't try to double check'.
2)Silicon lottery is throwing in a wrench, as usual.
3)The memory controller is 'fresh'.
4)Overclocks don't last forever, degrading over time. The memory controller inside the core can't do what it used to, either requiring a downclock or more voltage for stability.
-The memory voltage is fixed from vbios though, so that only leaves downclocking frequency.
-Once degradation starts, it's a slow trek downhill with you downclocking to a stable frequency, playing 'till it becomes unstable again, and repeating.
-The harder they're pushed, the faster the settings break. The miner friend would've already overclocked the memory for max profit. How they handled it is a completely unknown variable It could've been light, heavy, or something in between.

I suggest leaving memory clock at default if you intend to keep it.
Between games and mining, they stress different parts of the gpu. Power delivery/VRM for games, memory controller for mining. If either one goes out, you're out of a gpu regardless.

Should I maybe replace all the thermal pads and paste? Maybe the old owner messed it up or something when cleaning it? or applied the paste poorly?

If components aren't sustaining high temperatures, doing that will do nothing for the current scenario.


FYI: The voltage slider does absolutely nothing. It's more like a suggestion to the gpu that it is allowed to use more voltage, but Gpu Boost typically ignores it.

 

[Cookiezaddict]

You should stop wondering as you already know it was used for mining crypto already. Like a taxi, a 2 years old taxi will have 500% more wear and tear than normal people car at the same age. Basically that gpu was tortured its entire life!

Maybe you can RMA it LOL.

 

[Karadjgne]

It's a trade off. Think about the miners perspective, a decent one anyway. Gpus are not cheap. Better gpus are even more not cheap. So the miner is 'in the red' from the start. Then he's got the power requirements and that bill.

Best case scenario right now, it's taking a miner @ 500+ days solid mining, just to break even. The only way to shorten that gap is to undervolt the card as much as possible and remain stable because the gains from OC mining do not offset the power use or costs.

So most serious miners aren't jacking memory clocks and running the risks of trashing the card, having glitches, bit-flips etc. They are undervolting to bring power use down and still stay stable. Build-zoid has had several mining gpus that showed no signs of abuse damage, and has done several vids on the subject.

But that's a miner in for the long haul money. That's not a wannabe miner thinking about jacking up the card to get rich quicker.

OC is OC, there's no guarantee of any limits, either minimums or maximums, you get what you get according to the Silicon Lottery. Any degradation falls under the umbrella of Murphys Law, if it's going to happen, it will happen, it's just a matter of When.

 

[USAFRet]

Return it to absolute basic stock settings.
Voltage, freq...everything.

Does it run?
If so, leave it like that.

 

[realflow100]

my friend got 4 of the same gpu all used to mine and he kept the best one lol. and gave me the "middle" one of the remaining 3
One of them was so bad it crashed at +90 core +750 mem and was still not stable.

his best one did 2100mhz+ core clock almost always in games and benchmarks. and +1000mhz memory or higher.
I guess the one he got was lucky or a "fresh-er" gpu or something.

ive decided on +137mhz core +200mhz memory for now since memory overclocking gives almost no FPS benefit.
Technically +137mhz is the same as +140 so I guess I can just enter +137mhz

 

[Cookiezaddict]

my friend got 4 of the same gpu all used to mine and he kept the best one lol. and gave me the "middle" one of the remaining 3
One of them was so bad it crashed at +90 core +750 mem and was still not stable.

his best one did 2100mhz+ core clock almost always in games and benchmarks. and +1000mhz memory or higher.
I guess the one he got was lucky or a "fresh-er" gpu or something.

ive decided on +137mhz core +200mhz memory for now since memory overclocking gives almost no FPS benefit.
Technically +137mhz is the same as +140 so I guess I can just enter +137mhz

Dude it's free for a reason, if it's still great may as well sell it for $150 than giving it for free. Enjoy what you got and if I were you, I'd actually just run it stock since the card must be worn out already, being able to enjoy free card for as long as possible is great when it's actually something you cannot afford at the moment. If you can afford buying a card? Then sell it on ebay for $150, many would even pay that even though it can only run stock. and you can give yourself a used not ex mining 3060 ti that's still cost $250-300 in average in the US

Afterall, hats off for your friend who's giving you a GPU for free! It's legendary friend really, ya don't need to explain how he keeps the good one, lucky he gave it to you, if I were him I'd sell it on Ebay for $100.

 

[realflow100]

yeah its pretty awesome I got it as a few days early birthday gift its a pretty awesome upgrade over my 1650 I had before.
The amount of vram is tremendous.
I did some more testing and memory overclocking gave such petty gains I just decided to put my memory freq at stock.
Core clock can stay at 137mhz though. that gives a meaningful FPS boost. Not massive but a nice FPS increase. noticable.

 

[ohio_buckeye]

Compared to a 1650 the 3060 is a nice upgrade. I’d suggest set it all to stock and enjoy the card as is. You know your card was run hard. Don’t finish it off.

 

[Cookiezaddict]

Compared to a 1650 the 3060 is a nice upgrade. I’d suggest set it all to stock and enjoy the card as is. You know your card was run hard. Don’t finish it off.

That card now is like nearly worn out, if running stock may have another good 5 more years if keep getting tweaked, stressed tested, tortured every now and then, let alone boosted and boosted and boosted, I wouldn’t surprise if it breaks after a year or two.

more advise to run stock but in the end, OP can do whatever he wanted haha

 

[realflow100]

memory clock is fine being left at stock since theres like margin of error increase in FPS when only being able to raise it to like 300mhz lol
core clock is still ok though at 137.

 

[realflow100]

I'm planning on replacing the pads and paste on it.
I'm pretty sure the old ones are totally worn out. My friend has same card and it runs cooler with higher ambient temperature and lower fan speed! And even overclocked higher too compared to mine.
And mines still running hotter.
69-71*C when gaming with fans on max.

 

[KyaraM]

Honestly, if it runs fine at stock, I would just run it stock. As stated before, it's a mining card and it's kinda a small miracle it doesn't have issues at stock at all. Most aren't so lucky. I doubt you would feel much difference anyways, in most games it shouldn't make a huge difference between OC or not. But if you want to run it OC'd, feel free, just be aware that it might turn unstable after a while forcing you to adjust. Btw, did you try undervolting? Maybe that could help you as well.

 

[realflow100]

undervolting just led me to lower overclock in general.
Its quite difficult to find stable point when undervolting. because its jumping around so much. and will randomly become unstable.

For normal offset overclocking theres a very precise point where it goes from 24/7 stable to obviously unstable between +137mhz and +150mhz core clock offset
Heaven benchmark was the best for finding instability out of the games and stress tests I tried.
Within a couple minutes at +150mhz crash in heaven
but fine 24/7 at 137mhz (140mhz is same as 137mhz)

Usually when gaming I'll see clock well above 2000mhz and sometimes touching 2100mhz for a short time.

It usually starts out at 2145mhz to 2100-something then gradually lowers until the temperature stabilizes. to just under 2100mhz. between 2030-2095mhz or so.

upping the power limit and getting lower temps is the best way to get more mhz core clock.
I'm thinking of making a really crude/cheap chiller setup just using part of the air from my window air conditioner and redirecting it with a hose to my pc. that will significantly lower my gpu temps very easily. and super cheap. which also increases clock frequency.
gpu draws less watts at same volts so it can boost higher. due to less temperature.

 

[hotaru.hino]

Here's a question: is there enough performance between the stock settings and overclocking for this to be worth spending time and money pursuing? Are you really going to notice the performance difference? Think about it, 2100MHz is only 5% better, and that doesn't actually translate to 5% performance in games.

gpu draws less watts at same volts so it can boost higher. due to less temperature.

It doesn't work like that. Temperature is a function of heat dissipated from the cooling system vs heat produced. It doesn't mean the GPU is drawing less power.

 

[ohio_buckeye]

Trust me, the op is going to do what they want. We've been trying to tell them to leave well enough alone not to push their gpu anymore so they don't have issues....

 

[Karadjgne]

Power = volts x amps. If Power is a constant, lowering voltages means a rise in amperage. So if the card wants 300w and its at 1.4v, that's 214amps from the VRM's. Drop that to 1.3v and that's 230amps from the VRM's.

The heat just isn't as evident in the gpu itself, but can and will affect the VRM's temps, which isn't generally reported, and can affect stability depending on the power phases on the card and tolerances of the VRM's. If the card has 6 power phases running 60A mosfets, no worries at 230w. Same card running 40A mosfets and 214A is far more reliable than 230A and far less likely to burn out a slightly less than perfect mosfet.

With as tight as performance gaps are now with cpus and gpus, the benefits of OC are all but extinct. There's simply not enough gain to warrant the expenditure

 

[realflow100]

How does the gpu get away with boosting higher and higher with lower temperature. At the same power limit??
The voltage goes up and frequency goes up higher boosting on the voltage/frequency curve. regardless of power limiting status. as long as the temperature goes down. it boosts higher. at the same power/wattage limit?
How? Thats really weird to me.

 

[hotaru.hino]

How does the gpu get away with boosting higher and higher with lower temperature. At the same power limit??
The voltage goes up and frequency goes up higher boosting on the voltage/frequency curve. regardless of power limiting status. as long as the temperature goes down. it boosts higher. at the same power/wattage limit?
How? Thats really weird to me.

A GPU will boost until it hits one of these limits: temperature, what voltage it considers reliable, what voltage it will never exceed, and power. If it's not hitting one, it'll continue to boost.

You can check why the GPU isn't boosting further by using HWiNFO under "GPU Performance Limiters" or GPU-Z under the Sensors Tab -> PerfCap Reason

 

[realflow100]

Power limit. it just says power limit.
but it boosts higher and higher regardless of perfcap reason. if the temperature is lower. it just "does" boost higher anyways. try it yourself and see. load up furmark or something. and watch the clock start out high. then start dropping more and more. as the graphics card heats up.
crank up the fans to lower the temperature. and it boosts HIGHER even with power limit as the perfcap reason!
HOW is it doing that without using more wattage?

 

[hotaru.hino]

Power limit. it just says power limit.
but it boosts higher and higher regardless of perfcap reason. if the temperature is lower. it just "does" boost higher anyways. try it yourself and see. load up furmark or something. and watch the clock start out high. then start dropping more and more. as the graphics card heats up.
crank up the fans to lower the temperature. and it boosts HIGHER even with power limit as the perfcap reason!
HOW is it doing that without using more wattage?

I've seen this effect happen before, but I didn't look too deeply into it. The only explanation I could think of was that increasing the cooling around the card also cooled off the VRM better. A hotter VRM doesn't produce power as efficiently, so it could be claiming that it was hitting a power wall even though it wasn't really hitting the power limit that I set it to.

Although https://www.techpowerup.com/forums/...fcap-pwr-below-90-pf-tdp.287299/#post-4617104 explains that performance limiting due to power is more complicated than "card is hitting 100% TDP"