Ok,
@citay I'm back. After dozens of experiments with bios settings and tests I ended up with the following: all P-cores are fixed at 55, all E-cores at 43, AVX2 offset - munus 3, XMP - at 5600 (my memory nominal frequency which by default sets to 4400), Memory Extension Mode set to Performance Mode, CPU Core Voltage mode set to offset mode, the offset mode itself - to minus by CPU, and the voltage itself offset to minus 0.09. Then Pl 1-2 - 253 watts, the current limit - to 400A, Lite load is auto and it shows Mode 9 (which is by default on my motherboard).
With these settings I could get OCCT Power test fully stable at 87 degrees, and the Cinebench gave me 35194 pts with the Bclk 100 MHz bios setting enabled (it actually levelled up the cores fraquency to a higher level - to 5500 Hz from the normal 5480-something), and I think I'll get about the same score with this bios setting disabled - cannot check because right during the testing the electricity went off. Waiting...
What's not to like, power draw is lower, temps are lower, score is higher... nice. Cinebench doesn't care much about RAM speed, but of course, enabling XMP is a no-brainer (not that it was disabled before), DDR5-4400 would be a bottleneck. Your fan curves are still not that optimized, it seems: In idle they're all well over 1000 RPM, i run mine with below 500 RPM there.
In the screenshot from Cinebench below you'll see 2 scores: the higher is with the Bclk 100 MHz On the lower - OFF. HFInfo taken with Off.
Yes, i wish i could set my BCLK/FSB to 100 MHz on the dot, but my Z690 board's BIOS doesn't expose this option, seems to be only on Z790 boards. Spread spectrum is mainly good for passing EMI/EMC testing, for normal use i'd like to use fixed 100.0 MHz.
I know Intel recommends those mOhms be equal to each other, and the Intel Profile set it to 1.100 by 1.1000, but the lite load modes always set different values there, the first being lower than the second.
Note that in the
latest recommendations, there is no mention of having to set the same values for AC and DC loadline anymore, unlike in the
initial recommendations. It probably just added to the confusion, or they communicate this only to the board makers now.
The problem is, the only way to truly set everything correctly for a specific board model's VRM is to use the Intel VRTT (Voltage Regulator Test Tool). It's a huge tower of hardware that plops onto the socket and can be programmed with loads of up to 600W continuous and 800W peak, all the while capturing waveforms and data of what is going on, and if overshoot/undershoot etc. all stay in line with the specifications and don't deviate too much.
Here is an Intel document on this VRTT used for
FPGA testing,
https://cdrdv2-public.intel.com/632778/an853.pdf
Here's it being used on LGA1700,
Google-translated,
source
For an end user, you never really know which values are truly the "correct" ones. As you can see from your own result, you can optimize a lot about an individual CPU, lowering the power draw, increasing efficiency, avoiding thermal throttling and excessive voltages and currents. But to get to better operating parameters for all users, in light of these instability/deterioration issues, without them having to manually optimize everything (which can be too much to ask for a lot of users), this will have to be done by the board makers via BIOS updates, which are still in a state of flux.
