[SOLVED] AIO questions/issues - Fractal Celsius+ S36 Prisma

[S13ontap]

So I've been trying to track down the cause of some issues that I've been having with the system I built. I have the aforementioned AIO, but I feel like it's running hotter than it should, or not cooling as well as it should... Does anybody know how to troubleshoot this kind of stuff? Maybe I just have something wonky, like being plugged into the wrong port - AIO in cpufan vs pump. Wrong settings for PWM vs auto on the pump/bios settings. Does anyone know how to test if pump is functioning properly? why it maxes the fans all when I switch it in off auto fan control to PWM in the bios... etc....

There is not a lot on these things in the way of issues or reviews, I'm wondering if someone here gas some general knowledge bombs they can drop before I toss this thing in a lake....

 

[S13ontap]

LLC is line level calibration. The cpu doesn't really work at a constant rate, there's very slight dips when it's demand is next to nothing, then it starts a new line, and has a gap, new line etc. It's those gaps in demand that have limited voltage, are the vdroop. But the cpu has to start work before it can demand enough voltage to cover the work. That's where you get instability, not enough supplied voltage to cover what the cpu needs, before it demands it. That's why LLC was created. It's a pre-emptive voltage, an addition to raise the vdroop up high enough to bridge the gap.

BUT, it's an added voltage, so also applies to the top of the demand too. For instance, cpu hits the gap in need, demand drops to 0.9v, starts a new line and needs 1.2v, but will demand 1.4v just in case. Being at 0.9v, for that instant its unstable, bluescreen. LLC can add 0.3v to the droop, maintaining the 1.2v necessary and the cpu is happy. But it also adds 0.3v to the demand, so the cpu gets 1.7v and your temps are not happy at all.

LLC should be used judiciously, for most there's never a need for 6 or extreme or 100% or you end up with a cpu using 1.7v or more. Not good for the cpu at all, and ppl can't get stable because they keep dropping vcore to avoid the 1.7v they just saw in Hwmonitor. Most OC will never need more than 3-4, medium, 55-66%.

Theres vcore, vid, cpu voltage and other names used, but what's important is figuring out if that's actual cpu need, cpu demand, cpu supplied and whether it's after LLC adjustment or before. And it's different with every software, every bios. Gotta learn your particular bios settings. Getting all that in balance makes for a successful OC at voltages the cpu can live with and temps you can deal with.

Thanks! I had a loose understanding of LLC, and that just confirmed, as well as expanded on my understanding of the sytem.
As for the 4 setting, I got that from Silicone Lottery's recommendations, as well as what appears to be the generally accepted setting for best results. Dialed it in, and seems to be a bit happier.
Still under 80 for Thermal compliance, so it seems like I'm still ok.

From your explanation, I would expect to see higher fluctuation with the start/stop tasks that I could simulate in a Real bench or cinebench, correct?
I don't see the same peaks in P95, as would be expected from a relatively stable test. From my understanding, those stability testing programs are better at re-creating the peaks and valleys of real use. I expect that I may see higher temps there.
Sound right?

 

[Karadjgne]

P95 is a constant load, so there's not much variation, only when it's hitting the gaps in code. Asus RealBench runs multiple different aps to varying degrees, so can have much more pronounced vdroop periods.

Most ppl try to avoid bluescreens, but when testing stability, you actually take a proactive attempt to force a bluescreen situation. The difference between a successful OC and a failure is dialing in the voltages to account for the possibility of a bluescreen and bypassing the situation. Is why stock voltages are so high to start with, accounting for every variable in every cpu.