What are the missing variables that demonstrate the actual causation between cooling and results?

[londiste]

I am surprised that those who designed the motherboards made the extremely illogical decision to put the M.2 slots directly under the hottest component in the PC case - the GPU.

Heat-wise yes. But it is not illogical at all.

The location is dictated by need to run the PCIe traces from CPU to the slot. Yes, it can be moved elsewhere - and in some cases has been moved - but this is an extra effort and needs more care to still keep this stable. I bet this is considerably worse with current gen when that first M.2 slot is PCIe 5.0.

Other directions are already more occupied - RAM with more traces and more difficult requirements and VRM which is potentially even worse due to running power from there to CPU plus heat. And standard locations of things do not really allow putting stuff at the top (and mostly there are parts of VRM there as well). Also, M.2 is supposed to use <10W and motherboards mostly have a heatsink over there that is often connected to some larger heatsink.

In my own testing, and in the testing of some other people, I've noticed anomalies where the data we're tracking does not reflect upon the CPU/GPU temperatures, nor does it correlate with the clock speeds.

This means that we're not tracking the right things in HWInfo or whatever. Yet, there are so many options and I do not know what else to consider. I track averages and maxes for (DTS and Enhanced) CPU core temps and CPU package; GPU hot spot, memory junction and temperature; average effective clocks, max CPU/thread usage, and total CPU usage; motherboard and VRM temps, and the fan speeds.
...
Edit: TO CLARIFY: I need to know which sensors to monitor to discover the causal relationship between cooling and performance. None of the above are doing that, nor are the ones I've been watching, nor does watts.

This is far from trivial question. You are tracking (some) right things but the correlation here is complex and with all the feedbacks it gets worse. If you want to get some understanding into what the effects of parameters are start making some of the things constants. The principle is - everything matters.

Not claiming to be completely correct and some of it is rant-y but some semi-formulated thoughts.

The physically easier side is the cooling.

You dump x W of heat into it and it gets dissipated through some rads with help of fans and whatnot. For things to be comparable you need to use the same case, same case fans and their configuration. Ambient temperatures obviously matter. Fan speeds matter. In addition to fan speeds AIOs have pumps, their speed matters. Mount the cooler in as similar way as possible - for air coolers whether the air is blown up or towards the back matters. For AIOs where you mount it matters, a lot - mounting AIO at the top is convenient and nice but this will inevitably take warmer air from inside the case compared to another AIO mounted for example to the bottom of the case where it can suck in cooler air from outside.

Here is the first tricky bit - fan and pump speeds matter. As soon as you let these be dynamic, good luck figuring out how the effect correlates to speeds - it is going to be in a curve and possibly per fan or speed domain. Also there might be strange effects from how air moves, when things start to overheat etc. So, to keep things comparable, fix the fan speeds (all fans, including case fans). Measure in a couple predetermined speed points if necessary.

Trickier part today is the consumer side - CPU (or GPU) itself. In part because there are a number of dynamic things and in part because with contemporary hardware you may not have control over that.

Cooler does not care what the frequency is, or voltage, or usage. Cooler takes the watts in and dissipates the heat to the best of its ability, the result is CPU temperature which CPU cares about. Throttling is the obvious and simple one - if cooler cannot dissipate heat dumped into it, CPU temperature will go over threshold and it'll throttle. Other things are trickier. Attainable frequency is related to voltage which as the end result gets the W put out from CPU. frequency-voltage relationship is on a curve. Voltage to W is not usually linear either. So far so good - frequency shouldn't depend on heat directly outside throttling... but does due to secondary effects like heating up VRM due to power conversion waste or simply rising temperature inside the case.

This is assuming you can get the CPU to try and be at a certain frequency and do a fixed amount of work. But, frequencies are dynamic today. Frequencies are lowered if some limit is hit - temperature, various power limits, voltage changes (or voltage reliability changes). Some can be monitored, others less so. Best practical bet is probably something like you already have done - run something heavy and constant like Prime95. Making anything possible in UEFI into constant helps - set fixed frequency that can be held regardless of load, not allow any voltage drifts etc.

And the end result... is that the test does not reflect your actual usage

 

[ShrimpBrime]

I have the Asus TG H670-Pro Wifi D4. No clue about the default. I haven't messed with that setting at all.

I'll stick with HWiNFO64 if GPU-Z only gives me the package temp. Seriously, dude, I'm looking for causality, not one sensor. That I've already seen again and again isn't tied to causality by itself - not by a long shot.

Here's another question for you: when it comes to my mainboard, are the enhanced CPU package and core accurate or not in HWiNFO64? There's some info in the help file that says it tends to be inaccurate...Which is why I use DTS.

Yes I know gpuz has 1 cpu sensor. But it's light weight on the system while HWInfo64 is polling from all available board sensors and polling sensor that don't even exist on the board.

Just monitoring skews your results is all I was getting at there. This must be taken into account.

It's not accurate. The polling rate by default is 1000ms. So what you just saw, happened a minimum of 1000ms ago. This holds true for all temp programs. It's not even close to real time.

So you do a bunch of testing, come out with averages.

I'm not familiar with your board. My Z690 and Z790 boards seem to change the LLC if I set XMP for example. Default seems to be the lowest LLC setting. XMP enabled, the setting changes to a higher LLC. Sometimes 3, sometimes 4. Might be something to pay attention to. Again, it's simply an idea. A suggestion. Nothing more.

 

[:D:D]

So how is it that a 280 "won" over a 360 of the same brand and product line (minus lights)

The 280 was sucking room air from outside the case while the 360 was sucking heated air from in the case. Perhaps some temperature sensors for measuring inlet temps would be useful.

 

[TPCEA]

The 280 was sucking room air from outside the case while the 360 was sucking heated air from in the case. Perhaps some temperature sensors for measuring inlet temps would be useful.

A1 and E1 had the same airflow configuration.

 

[:D:D]

A1 and E1 had the same airflow configuration.

IIRC the guy said 31C room temp so side rad is using that air temp to cool the 280 rad while the top 360 uses internal air which is maybe 5 or more degrees higher going by the MB temp.

 

[TPCEA]

Looking at the details of how he chose to test the 2 sizes of AIOs, there are confounding factors, such as:

• Using glass on top for the 360s, but mesh on the side for the 280s,
• Mounting the 360s on top but not the 280s,

On the other hand, he has produced this video to test the case, not the AIOs, so he even swapped out the Arctic fans for Montech fans (which I think was a useless decision). Since his room temp is @31C, so this is actually extremely useful for millions of people in certain parts of the world who, like him, cannot control their room temp. Those people need the best choice in cases to minimize temp problems.

All of these factors can explain the strange things I've noted, and demonstrates just how hard it is to do testing. I know some people prefer lab results, but those can sometimes be vastly different from real-world results. The fact is that there just isn't any way to produce results that are reproducible for most people, so it is unrealistic to expect that every tester will get the same results, especially when we consider that there are billions of trillions of possible configurations..

 

[redeye]

… advertising …

 

[TPCEA]

… advertising …

What do you mean???

 

[redeye]

What do you mean???

advertising (through the ages) makes you expect better than it delivers…

AIO’s can not be made to be better than than they are, unlike Nvidia GPU’s with DLSS. /o

 

[TPCEA]

Agree on the first point, although it isn't actually relevant to my topic.
Disagree on the 2nd. That has repeatedly been done or the current best AIOs would not be any better than the previous ones. Also not relevant to my topic.

 

[redeye]

Agree on the first point, although it isn't actually relevant to my topic.
Disagree on the 2nd. That has repeatedly been done or the current best AIOs would not be any better than the previous ones. Also not relevant to my topic.

So, OP your question has rings of “please write my thesis for me…” lol

because that what a thesis would answer…

 

[TPCEA]

So, OP your question has rings of “please write my thesis for me…” lol

because that what a thesis would answer…

I'm 57, kid, and I'm never going back to Uni. You're welcome to absent yourself from the discussion if you're going to be rude.

 

[bug]

Here's a sneak peek into how transistors really work: https://www.electronics-tutorials.ws/transistor/tran_7.html

Simplifying a bit, temperature doesn't govern how they work at all, voltage does. Usually, if you can cool them better, you can apply more voltage and that will make them switch faster. But, as you may have guessed by now, the quality of the applied voltage also matters (quality of PSU, VRMs and whatnot).

I'm 57, kid, and I'm never going back to Uni. You're welcome to absent yourself from the discussion if you're going to be rude.

I don't think he was trying to be rude. He was just pointing out the proper answer to this can earn you a degree in electrical engineering.

 

[TPCEA]

Here's a sneak peek into how transistors really work: https://www.electronics-tutorials.ws/transistor/tran_7.html

Simplifying a bit, temperature doesn't govern how they work at all, voltage does. Usually, if you can cool them better, you can apply more voltage and that will make them switch faster. But, as you may have guessed by now, the quality of the applied voltage also matters (quality of PSU, VRMs and whatnot).


I don't think he was trying to be rude. He was just pointing out the proper answer to this can earn you a degree in electrical engineering.

Well, then, thanks for translating.

So, if the "purity" (so to speak) of the electricity is critical for transistors, then would that suggest that ratings from Cybenetic Labs would be more helpful than 80+?

I would guess that redundant circuits, voltage regulators, etc, would be part of what makes a PSU great. Do you know of resources that are trustworthy in rating PSUs?

I guess that this means that it would also be better to use cables, connectors, RAM, mobos and so on that have better quality components and materials, too?

 

[bug]

Well, then, thanks for translating.

So, if the "purity" (so to speak) of the electricity is critical for transistors, then would that suggest that ratings from Cybenetic Labs would be more helpful than 80+?

I would guess that redundant circuits, voltage regulators, etc, would be part of what makes a PSU great. Do you know of resources that are trustworthy in rating PSUs?

I guess that this means that it would also be better to use cables, connectors, RAM, mobos and so on that have better quality components and materials, too?

Sadly, I'm not too familiar with any of those. But yes, what you are looking for is clean, stable voltage. It's why high-end mobos use more VRMs (but again, that's simplifying things, VRMs need to be of high-quality themselves, adequately cooled and so on).

 

[:D:D]

Here's a sneak peek into how transistors really work: https://www.electronics-tutorials.ws/transistor/tran_7.html

Simplifying a bit, temperature doesn't govern how they work at all, voltage does.

FinFETs are a little different to MOSFETs.

What about temperature inversion effect? Seem to remember Intel saying VID is adjusted with temperature too but don't know how much the latest tech is affected.

 

[redeye]

I'm 57, kid, and I'm never going back to Uni. You're welcome to absent yourself from the discussion if you're going to be rude.

really?