Temps good, PSU sufficient but a bit of a glass ceiling with my OC

[NewbieOneKenobi]

Rig:
CPU: i5-9600K
mobo: Asus Z390 Aorus Pro
mem: 2x8GB Ripjaws V 3000/14
GPU: Colorful iGame GeForce GTX 1070Ti Vulcan U Top
system disk: PNY CS3030
other: SB Titanium PCIE
case: Phanteks Enthoo Pro

Power supply:
PSU: Corsair RM750W (yellow label, gold-rated, 61A on 12V)
Mobo has an additional 4-pin EPS socket, which I can optionally plug into 6-pin PCIE through an adapter. PSU only has 2x(8+6) PCIE and 1x8-pin EPS, so I plug both PCIE 8s into the GPU (it's 8+8, instead of the usual 8+6 for a 1070ti), which means one of the cables has plugs going into both CPU and GPU.
CPU rarely gets to 170W as per HWMonitor.

Cooling:
CPU: TC14PE with 2xP14 on the sides (CPU_OPT) and SW3HS 14 centre (CPU), Cryonaut (dotted x method)
GPU: Accelero Xtreme III (no fan mod but thinking about it)
Air intake: 2xP14 front, 1xP14 bottom
Air exhaust: 1xSW3HS 14 rear, 1xPhanteks 14-something top

Current OC:
CPU: 4.7 @ 1.4 (showing 1.392 most of the time) but with boost enabled and its multiplier changed by +3 ('CPU Upgrade: Advanced' in mobo settings)
RAM: 1.4 just to be sure (1.35 XMP)
GPU: 2100/9100, PL 120%, voltage +53%

BIOS settings: disabled all C-States, VTd, etc., and everything else the official Aorus OC manual says should be disabled. LLC maxed out ('Ultra Extreme' or something like that) just in case.

Last TimeSpy score: overall 7475, CPU 6480, GPU 7684
Valley: 4500-ish @ 1440p ultra, AAx2 (from 3900-ish stock)

Temps aren't really the problem. I've seen up to about 92C on 5.1 attempts in Prime95 small FFT (passed but yielded lower TimeSpy scores than more modest OCs). I rarely hear the fans, even in benchmarks, forget in games. Running the fans at 100% doesn't actually change much compared to my silence-friendly custom curve.

Overdoing the OC simply results in failed benchmarks and bluescreens without temp issues. For example even when the GPU fails Valley, its temp doesn't exceed 65C. When the CPU fails Prime95, it's usually one of the cores crapping out immediately. Increasing the voltage by a large margin tends to help, but the results are disappointing (see above — lower benchmark scores than with a more modest OC) and the price too high. Even when I'm seeing 92C, the CPU doesn't approach 200W, so it's far below the cooler's probable TDP rating, whatever that would be. It feels like hitting the dies' limits, though I'm skeptical about there not being anything else one could do to squeeze out some more.

One curious thing I've noticed is that increasing the OC on the CPU tends to improve CPU benchmark results but harm GPU benchmark results. So… insufficient 12V line? Bus problems?

The 4.7 CPU clock with XMP on and boost enabled (with multiplier increased by a further 3) produces the best score of all combinations I could come up with (including higher stable CPU OCs with XMP disabled or 4.7 with no boost).

The GPU voltage setting doesn't do much, but it does increase Valley scores marginally up to a point, failing if too high (like 100% will fail where 100% passes) or too low (higher OCs seem to require more voltage to avoid crapping out, or alternatively they produce lower scores).

I'm not sure if the 65C temp point does something to the card, though I've never seen Valley exceed 64C, so perhaps there's a hint. However, the card had exceeded 70C on its factory cooler without crapping out.

Is there anything I could try? Perhaps some of those completely arcane detailed power settings on the last tier of the Aorus's advanced tabs? (Internal VR-something, yadda yadda.) Or something simpler that I may be missing? Or is the PSU my bottleneck? Or have I truly hit the dies' limits?

 

[Joxeon]

Have you overclocked the cache as that usually results in a gaming boost?

 

[NewbieOneKenobi]

Have you overclocked the cache as that usually results in a gaming boost?

Pushed ring from 43 to 47 after your post and got a small bump — about 150 worth of TimeSpy CPU Score. Thanks!

***

Right now, I'm also running the memory at XMP but raised to 3300. I've figured out by testing that leaving Enhanced Multicore Performance at Auto is better than either enabling or disabling it (which either crashes the benchmarks or lowers the score at my current settings, respectively).

What's worrying is that supplying 1.4V at this configuration results in almost 100C temperatures despite all three CPU fans running at their max RPM (ca. 1620 for the SW3 and ca. 1720 for the P14), which looks puzzling to me, considering that the CPU's output hovers just below 180W, according to HWMonitor, which seems far less than my cooler should be able to handle. Makes me ask how e.g. Le Grand Marcho RT is supposed to provide sufficient cooling for >300W, or whether a D15 with standard standards wouldn't be doing a better job than my Frankenstein. Trying 1.5V — which is where Auto got me — resulted in >100C temperatures. Trying 1.3V resulted in a freeze crash in Prime95. At 1.35V set, I think the mobo ups it to just below 1.4 anyway, which one of the six cores reaching 98C and one or two more entering 95C zone, which I currently have set as the alarm value.

I may simply have poor luck in the lottery, but something in me is approaching this with disbelief. We're looking at three CPU fans and five case fans running at their full RPM struggling with a 9600kf needing 1.4V for stability @4.7 GHz, which is not even on all cores and is cooled by a freaking TC14PE with three fans higher-performing than its stock fans, in a well-ventilated case. And did I mention a relatively high-end mobo with a beefy power section.

This is a bit of a demoralizing experience — kind of making me doubt the value of K processors and even more so Z chipsets, and beefy coolers, and higher electricity bills, as opposed to skimming the mobo budget and the cooler budget, and the electricity budget too, buying a correspondingly more expensive processor and running it at stock.