wow its a big difference
 
usually i can run the card +210 core on becnhmarks and +175 on games and stable but your difference is alot
 

When I'm in game highest Temp was 55C ,
are you using Classified Tool ?

What bios are you using? Are you running +1650 mem in games? Lower that to +1000 or +800

Probably as expected.  Kingpin is already factory overclocked so to say a +500 or +100 OC is already on top of an OC.  The 3090 double sided memory design combined with the hybrid cooler makes the card a different kind of challenge to keep the heat away during extended use at peak clocks.  I've had to turn down my OCs for gaming stability even though I can run short benchmarks at a more aggressive OC all day.  Plus things like if the game is rasterized only vs RTX can also challenge the stability of an OC.
 
Nice score in PR btw.

On stock bios I can achieve Port Royal score of 15110 and consistently low 15k with 133 on core and 1350 on memory. This is with the card in case and cooler outside of case. Anything above those clocks and it will crash in Port royal.
With ambient temperature of 28 celcius with card and cooler in case my score drops to 14750. Im using the h500p mesh case and 360 gpu rad mounted on top of case.

Keep in mind, Kingpin dies are pre-screened, if you can't maintain at least 2100MHz rock stable on the core in games/benchmarks, something is definitely wrong, that's what they're pre-selected to be able to do bare minimum.  Most people I've seen are hitting ~2200MHz or a bit less before they actually have to start futzing with the Classified Tool, from what I understand.
My card will be here tomorrow, should be interesting.
The base clocks on the Kingpin are 120MHz higher than a 3090 FTW3, so keep that in mind when comparing offsets with people, your +75 is a higher clock than their +165 in actual boost clocks as a result.  As a reference for that, my 3090 FTW3 Hybrid, without any overclock/offset, hits 1905-1935MHz on the core - pretty sure Kingpins start off in the 2040-2070MHz range, right?
 

kraade
Game stability vs overclocking benchmarks are way different,  Benchmarks are a Drag race that only have to handle the load a few minutes, Overclocking for a game in which you load the GPU for over an hour  tweek your settings and its going to be like 200 Memory and 100 on the GPU clock,  Use MSI Kombuster and walk and check it every 20 minutes, Go away for 2 hours and see how your system is doing . then you have a stable overclock, this is the same thing if you have a 3090 KPE or a 1050ti...

 
And yes, this.  Games can be as or more stressful than some benchmarks, and you're running games for much longer than any benchmark.
If you want to ensure stability on a GPU OC for 24/7 use, you should be running it through the ringer, Port Royal, both Time Spys, even all 3 Fire Strikes if you can, and multi-hour gaming sessions with games that both do & do not have RT stuff, Cyberpunk 2077 is a great stress test if you have it, turn off the frame rate cap & just let it eat - if it crashes during any of that, it's not stable, simple as that.
On my 3090 FTW3 Hybrid, I run 2040MHz @ 1000mV (an overclock/undervolt) and get cooler temps, better than stock performance, and lower power draw, all in one package, it's quite nice. 

Offset OC is better for gaming because internal clocks are almost identical to the requested clocks. You haven’t mentioned your clocks and temperature while gaming.

I need some help as well. My best Port Royal is 14231 with the Stock Bios, 105% power limit, +140/1200. Tried switching to the OC bios and LN2 bios and neither of them game me higher scores. 

I will record power draw and come back to you ,Thanks

Yeah Cyberpunk was the worse for me. I thought I had it rock stable after gaming for hours, then I walk across the nomad camp and it crashes in 20 seconds.

Cyberpunk crashes a lot at the best of times. Don't use it to decide whether you're stable or not.