I would check your CPU block mount. 96C is pretty toasty if you are using 2 XE 480s. What kind of spread are you getting between cores?Thanks for taking a look JustinThyme! I'll walk through the steps I have been taking up until now and then include results of your suggested starting point:
My per-core overclocking methods
1) Set each individual core to 1.19
2) Start at 43x per core
3) Complete 20 minutes P95 26.6 non-avx Blend test + Aida64 Cache Stress 20 minutes
4) Bump up another multiplier for each core which stays below 70c (my target core temp when water reaches it's fully heat soaked temp of 32-33c) Stupid Southern California winters
5) Rinse repeat until a core reaches 70c OR there is BSOD/thread failure
6) Upon failure, add more voltage until stable or temps are reached
Now for the results of the 10 minute R23 run @ x47, 1.225v core, AVX Offset -0, LLC Level 4, CPU Current 200%, Mesh x28-31, and everything else on auto:
CPU Package max: 96c
PCH Core: 1.0v
VRM Temp max: 89c
Cache Voltage: 1.209v-1.216v
All in all, it looks like the AUTO options gave the system a lot of juice. I have never seen my PC draw 900w from the socket before from just a CPU stress test. The AVX offset of 0 was fun to test but i'm not sure how important it is if I don't use a ton of AVX instructions in my day to day. Can anyone recommend another stress test for non-avx workloads? Or is it such that the system is deemed stable for all instructions if AVX passes? The issue here is that I think P95 Blend is not the best stress test since it's limited to 16 cores / 32 threads. I would like to get above 47x all-core WITH an AVX offset.
Putting a fixed cache voltage of 1.1 instead of Auto should shave some degrees off the package too, but it'll jump right back up once I start tuning DRAM.
For VCCIN, I'd go for 1.82-1.85.If you are using HWInfo, check to see what your VCCIN drops to under load. My chip seems happy when the VCCIN voltage doesn't go below 1.75.
That is HOT for the VRMs. Maybe rig a fan over them or add a VRM water block to cool them down a bit.
The only way to "save power" on idle is to use adaptive voltage, which does not work too well for most of us due to the pre-programmed VID tables being so high. As an example, say you set your target adaptive voltage to be 1.225V with a 47 multiplier. If Intel's pre-prgrammed voltage for your chip at that multiplier is 1.32, that is what you will see. Your setting will be over ridden. I have tried this, as others have, and resorted back to a static voltage.
I tried the "By specific core" method and it took me ages and netted me no temperature improvements. I wouldn't waste too much time there. I would also not use the latest bios. The newer bios versions have security patches which really kill performance and may require more voltage to achieve a certain OC. I looked at the RVIE bios versions and I would try version 2002 which contains MC 29. You will have to use the bios flashback option to flash the bios. If the RVIE has 2 bios chips, you can always load 2002 onto the second chip and compare..