Sajin
Safe guards are still in place for ln2. If you want the xoc vbios you have to e-mail Vince.

xMAXxIMUSx
lol forget that then, sorry one more stupiud question , any way to unlock the mv slider on afterburner?

Chamidorix
I'm disappointed in you hyper and Sajin, just run TSE on 8K and it is super obvious it is clocking down due to power limit with average consumption way below 520W. I've spent about 20hours now messing with my KPE and there is simply no way to avoid it treating transient spikes to 520W as baseline for driver power throttling. Its utter nonsense we do all this power throttling talk on a 1440p benchmark that struggles to break 500W on 1.1V in the first place, on most non-garbage bin chips. Also, raytracing is merely the activation of a marginal extra set of silicon (BVH solve and tensor denoise) per SM, and most importantly a much deeper (not wider!!) pipeline per frame of compute on fp32. AKA, it does not increase power consumption in any way comparable to the widening the per frame compute load by increasing the render target (to 4k+ etc). 
 
While on a Strix or MSI card, the driver respects average power draw instead of transient spikes for limiting. So you can average 520W and spike to 580+ before Nvidia driver requests lower clocks. 
 
And it is not hitting the fbvddd, src, etc limit, those are pretty much ignored, to say nothing of even trusting the dumped values from BMJet in the first place. 
 
It is all extremely easy to observe with classified because you just lock the various voltages and observe how the driver reported values respond in GPU-Z or HWInfo - if you give it a ton of memory voltage you can get driver reported power on that rail as high as you want. While on the other side even if your curve is extremely conservative the second your nvddd rail is high enough to transiently spike to 520W once, you will downclock even though driver thinks you are at 0.8V or something silly. You will literally step down in clocks at the same reported driver voltage, while actually maintaining 1.1 nvddd via direct i2c/classified. 
 
The idea that you need to up voltages to make use of the power limit (implying that leakage is soo godly on the kingpin that it magically can't hit 520W average on 1.1V nvdd) stems from the fact that transient spikes are far far less with msvddd(cache) and fbvdd(vram), vs nvddd(SMs). So you can get your average power consumption up higher without the transient spikes dipping into 520W throttle territory as much, by just increasing msvdd/fbvdd voltages rather than nvddd. Then, also, obviously 1.1V doesn't hit very high current on Port Royale in the first place; again we should all be using a bench that actually loads more of the SMs like TSE.