AIO lifetime, swap or not while I'm changing CPU

[PineyCreek]

Bought a Fractal Celsius S36 late 2017 or early 2018.  it had a premature pump failure about 2-3 months later and they shipped a replacement to me.  That one's been running like a champ in my system since then (and still is).  Since it's coming up on 5 years and that's Fractal's warranty on this product, I'm wondering if I should consider swapping it as I'll have the system disconnected while I'm swapping CPUs.  Or, is it like most things and I should stop thinking about it until it actually fails or its cooling capabilities become substandard?  My first (well, technically second) AIO, so I'm curious if people view AIO warranty as the cutoff point for trusting it or not.

 

 

[aDoomGuy]

As long it works as intended I see no reason to replace it unless you actually want a new one. 

[Motifator]

This is a known to be kind of problematic AIO indeed, the pump was weak and it wasn't all that great even with decent performance on paper. I'd replace it personally.

[PineyCreek]

Ok, here I am post-swap with a few observations that I wanted to bounce off forum members.

 

I went ahead and swapped from 1700 to 5800X on the Asrock X370 Taichi board, everything went fine.  Now, before I make the below observations, I know the 5800X is a volcano of a CPU. Also, all temperatures stated/observed are Tctl, or the sensor on the CPU, not the sensor on the mainboard close to the CPU.

 

What I noticed is that the CPU is perfectly stable, but it won't boost to the rated 4.8GHz in CPU benchmarks due to it hitting Tctl of 90C.  I'm not too broken up about it because it never goes above high 70s in actual gaming or gaming benchmarks, and the CPU performance overall still feels about 50% better and the clocks are an easy increase of 25%.

 

I also noticed that the S36 lists itself as a DC AIO, yet has a PWM setting (selection ring) and a 4 pin connector.  It behaves as a DC pump/AIO as when you change it to PWM, everything maxes out.  Now, one would think that maxing out the fans and pump to where everything's easily 5x more audible would make a difference in thermal performance or readings, or perhaps prevent it from hitting Tctl of 90C.  It does not.  In fact, it makes no discernable difference whatsoever, so now I'm questioning if the age of the pump and/or fluid is catching up to me, hence impeding the thermal transfer between cold plate and radiator, or if this is just the way it is with a 5800X.  Right now I have the header set to PWM but the S36 is set to Auto with its selection ring.  Previously the header was set to DC and the S36 set to Auto.  I see no difference in temperatures, and the pump and fans are quiet. Setting both

 

Though, even on the quieter settings, I still see the boost in performance, I never hit Tctl/90C in gaming, and at current I can be perfectly happy as is.

 

What conclusions should I draw from this, and what if anything might I proactively do? Or perhaps the pump was weak from the start as @Motifator mentioned and now it's just more obvious with a beefier CPU?  It could also be that it's improperly seated and/or at an angle.  When in PWM mode on both the AIO selector ring and the header, it scales differently (as expected) but again, even at 100%, it can't prevent Tctl/90C from being reached with synthetic benchmarking, but again, I also never hit that temperature in any use case (thus far). Maybe I should find a almost 100% CPU-dependent game to test with.

[PineyCreek]

Well...tried a few CPU-oriented games that barely touch the GPU...seems the 70s is just where Tctl/Tdie likes to sit.  I can hit two core 4.8 (normal) or all core 4.4 GHz in the 70s...never goes to 90C on anything I've tested thus far...only on synthetic.  Guess this is normal, though I'd still appreciate your thoughts.

 

I'll note the sensor on the motherboard near the CPU behaves moderately close to the way it did when the 1700 did, running about the same at idle and only very slightly more than the 1700 did under load, as a point of comparison.

[freeagent]

On 11/1/2022 at 6:54 AM, PineyCreek said:

Or, is it like most things and I should stop thinking about it until it actually fails or its cooling capabilities become substandard?  My first (well, technically second) AIO, so I'm curious if people view AIO warranty as the cutoff point for trusting it or not.

Use the warranty after some time, don't let it lapse. The thing with CLC is they suffer from permeation. How long it will be good depends on how you use it, and with what CPU. Guys with custom loops top up every few months or so, how do you top up a CLC that is under warranty? I got about 14 months out of mine before I noticed the temps coming up. I retired it after 3 or 4 years as it couldn't handle my CPU at the time at stock towards the end.

[PineyCreek]

13 hours ago, freeagent said:

Use the warranty after some time, don't let it lapse. The thing with CLC is they suffer from permeation. How long it will be good depends on how you use it, and with what CPU. Guys with custom loops top up every few months or so, how do you top up a CLC that is under warranty? I got about 14 months out of mine before I noticed the temps coming up. I retired it after 3 or 4 years as it couldn't handle my CPU at the time at stock towards the end.

Oh, the warranty's already gone.  I checked and verified it was bought May 2017. The Celsius S36 came with a 5 year warranty.  To its credit, it seems to be performing as well as it did when it was first installed.  There's signs of permeation, sure, but for example, not enough to where you hear large amounts of sloshing, etc. when you move the case.  I've found out that my view of Tmax on synthetic benchmarks is flawed.  In the past few generations it's more common to ramp up high or hit max, and your performance is more based on how much cooling you have moreso.  For example, during those synthetic benchmarks when the internal CPU temp was always hitting its max limit, it still wasn't thermally throttling (per HWINFO64 flags and also frequency observations).