I have come to the conclusion that on the 1800X it is "optimal" to use at stock.
Once or twice I have posted in the C6H thread that I felt on the R7 1700 3.7GHz all cores OC was pretty sweet for voltage/cooling requirement and performance. At some point I would probably assess for my uses if 3.7GHz AC OC lost me anything against the 3.8GHz AC OC. Now on the R7 1700 stock AC is 3.2GHz, single core it will go max to 3.75GHz. So 3.7/3.8GHz OC got me a real nice boost over stock setup and I matched/beat max XFR single core.
Now on the 1800X gunning for 3.9GHz really pulls so much power from the wall when running some stress tests. I have noted worst case constant 300-350W for total rig inc screen when only CPU is under load. 3.9GHz needs VID: 1.425V on my sample, LLC: [Auto], cooling requirement under stress tests is also really at the edge for my air cooler IMO. I have noted with ambient temperature changes, that like stress tests will increase the max CPU temp, as the cooling really can't keep up with ambient air temp change. So basically WC is the route to improve that.
Now if I run it at stock I have 3.7GHz ACB at 1.25V MAX, which is similar to what R7 1700 use when OC'd to that level. It is so easy to cool, power usage is substantially lowered. I saw max ~250W in similar case tests as the 3.9GHz OC. Also running at stock the CPU does boost to 4.1GHz and at times between 3.7GHz to 3.9GHz on multiple cores (will be checking in more apps closely).
So am I just going for RAM OC/tweak only now on 1800X. I feel I should keep this vs the R7 1700 to run daily, as it's the best of both world.
First up some HCI Memtest.
I had picked arbitrary SOC: 1.05V and VDIMM: 1.375V as this seemed right for 3333MHz Fast on several of the CPUs I have used. I went for ProcODT: 53.3 ohms and VDDP: 900mV, with PLL/SB manually set as 1.8V/1.05V.
AIDA64 with 3333MHz Fast I have lost nothing.
Cinebench is also where I'd expect to be.
Next IBT AVX custom 13312MB GFlops, is ~180 at stock with 3333MHz Fast, 3.8/3333 Fast is ~185 and 3.9 / 3466 Stilt is ~188.
Power usage from the wall has dropped by upto 30%, temps are well even for when ambient raises, the quietness of the rig was astounding. Now one core did go to 4.05GHz, I will check log how long this was and when, but I believe for majority of the test CPU did not boost above 3.7GHz.
Now this same profile failed Y-Cruncher within 30min.
I noted a core went to 3.975GHz, after repeat testing it become clear that Y-Cruncher was a better test than IBT AVX custom 13312MB when at stock set up but RAM OC'd. As some cores were boosting above stock ACB of 3.7GHz which in IBT AVX they will not due to power draw, etc.
I am currently at ~10.5hrs of pass on 3333MHz Fast on Y-Cruncher, again power draw is lower, cores have been boosting up to 4.1GHz at times. SOC bumped to 1.1V, I felt this amount was needed as at stock cores will be boosting to 4.1GHz in other uses. ProcODT I went for 60ohms and also increased CAD bus from stock [Auto] of 24/24/24/24 to 30/30/30/30. I also bumped VDIMM to 1.39V again like SOC I have seen a tendency of this needing a bump with increased CPU speed and CPU oriented tests.
HWINFO is at a polling rate of 1000ms, AFAIK the CPU can change frequencies at 2ms, so logs will not really show the truest 'picture'.
All in all I reckon if OC'ing then 1700/1700X are best, as you have highest/higher chance of beating the maximum single core frequency at stock. On 1800X I reckon stock is best with RAM OC/tweak. I have done some wPrime and SuperPi, stock 1800X is so close to 3.9/3466MHz OC but the ease of setup for stock 1800X 3333MHz Fast is way easier and efficient on power usage, cooling requirement, temps, etc.