Just going to debunk this "CO" being an undervolt, it is not.
The DLDO is a Digital Low DropOut regulator, PSM stands for Pulse Skip Mode. CO aka Curve Optimizer sets DLDO PSM Margin... this tells the regulator how far it can overshoot or undershoot under load changes.
If you run your CPU at CO -30 and apply load, it will first undershoot because your margin is set low and then slowly increase to the point where it wants to run at given the power limits you set.
So for example, if you set your 5800X3D at 115/95/95 it will boost all the way up to 4450 on all cores and run it with CO 0... it will first overshoot its voltage and then lower it as the frequency changes. If you run it with -30 it will undershoot the voltage as the frequency changes and creep up to meet the power limit. In BOTH situations my (your) CPU will run at 1.216v... because that is what needs to stay at that load and it is allowed to reach that because 115 = 95 * 1.2... math never fails.
I also see a lot of false "overclocking" screenshots, where they show CPU-Z doing a multiplier * FCLK... this is SO WRONG. If you increase the FCLK past 100 your CPU will NOT increase its multiplier past 44. You CAN NOT overclock this CPU using a FCLK increase and assume it reaches the 44.5 multiplier in an all core load. It simply DOES NOT DO THAT, the SMU will NOT allow you to do it.
If anyone can prove me wrong, try me and show me a screenshot where you are running at that multiplier with your set FCLK. Also show me a CB23 result, because on average the 5800X3D my CPU scores 14.6k and I have yet to see a valid overclock claim score more than 15k... until then, everyone is full of it.
I have tried to fiddle with the SMU a LOT, if you write any value but 45.5 to the max multiplier register... your CPU will run at 3400 MHz. So this tells me there is a hard limit that the SMU controls, which is completely independent of used platform firmware.
The two screenshots below are from CB23, the first one is from last year and the second one is from today. These run are done on Ubuntu Linux, running CB23 in WINE and there are four VM's running in the background. Temperatures stay in the mid 70'ties. I have seen written testimonials where people claim to have reached over 15k, but again these are still stock and not overclocked at all.
Also regarding temperature, the measurements are made near the CPU cores themselves and there is a piece of insulation slapped on top. This cache sees a gradient of heat, but there does not seem to be anything to measure it. There is another sensor that measures the "package" or SoC temperature, under full load mine rarely reaches 40 degrees and when I touch my CPU block... that temperature seems to be about right as it is warm to the touch. Removing the load also confirms this, as it dives below 40 degrees immediately. So in my opinion, saying the heat density is greater is wrong. The measurement we read is simply made at a location where the most heat is... as it should. A 1g/cm2 coin at 100 degrees compared to a 2g/cm2 coin at 100 degrees equals 100 degrees regardless of density.
