Is there a formula to help normalize temperature testing when ambient is variable?

[TPCEA]

In my testing, I've suspected that (because of things like resistance and heat production, captive air in a case, turbulence, etc.) when the ambient temperature changes, so will the results, but it won't be linear, and now I've proven this to such an extent that I have to find a solution, if one exists, to normalize test results. Sadly, I have a few challenges in this regard.

I anticipate that there are people who are going to make suggestions, ask questions or try to derail this thread, so the following is a disclosure to head things off. See the bottom for what I'm actually asking for.

1. Mathematically, I'm no genius.
2. Financially, I'm unable to throw money at this problem.
3. Since there are a lot of smart people here, I thought I'd see if anyone can offer any useful help, i.e. formulas for temperatures of the CPU/iGPU (DTS and Enhanced from HWiNFO64), GPU, mobo and VRM, which are what I've been tracking of late for my AIO testing.

Thank you!

---

​

EDIT: Since the responses suggest that not everyone fully understands, which means I've not done my job in making everything clear, here's more. My apologies. I'm not asking for suggestions on the following information, merely trying to make things clear.

1. I am running tests not because I enjoy it or because I need to, but because I'm a reviewer-tester. At this time, I've completed testing the first AIO (TR Frozen Notte 360) for its final review.
   1. I always do an initial review based on published specs, and then at least one more review based on test results.
2. I'm also currently engaged in running a series of tests to determine the best configuration for a 360 AIO. Temperature variance has revealed that there is a non-linear progression in the temperatures, especially the CPU. I would link to the info, but I'm afraid I'd be banned again for linking to my own work, even though it would help. Suffice it to say that you can find it on BC under the name RevGAM.
3. I'm not having cooler problems.
4. I cannot control the temperature in my apartment, especially outside the range of my window AC. The best I can do to heat things up is a space heater next to my PC, which is highly imprecise.
5. I'm looking for formulas. Really.
6. It's not a matter of whether ambient variance is abnormal or not, but on the impact it has on my testing. To cite one example, if my ambient is 24C, the DTS CPU core temp will be 9C higher at 28C ambient, not 4C higher. Please don't explain why this happens - I understand. See #5.
7. My PC is in a bedroom, my thermostat is in the hallway. It's not relevant, however, since my heater isn't going to turn on when the apartment is warmer than 60F.
8. Again, I do not have an effective way to always keep the temperature the same but, once things warm up enough, my window AC might be useful for that.
9. I do not agree with the use of an open bench, nor do I agree with lab results. I want real-world results, not sterile results captured in devices and chambers that are the opposite of real-world. Very few (not none, but out of the billions that use a PC, it's about the same as none) people run their computer on an open bench.
10. With that in mind, opening up the case is also not a good idea, since a lot of people don't open up their computer and blow a fan in there, nor do many feel comfortable doing something so "terrifying", plus anyone with pets or small children would be foolish to leave their PCs open.
11. I run Prime 95 with Small FFTs for 1 hour, which I'm sure someone will say that it's more than long enough, so please don't. I do this to make sure that the rad is as heat-soaked as possible. To aid in this, I also watch a video and do some work at the same time. I then run Cinebench 2024 CPU Multi-Core while watching a video and, finally, I run Time Spy (with no other activities since it is sensitive). If I make a mistake, I redo all or part of the testing depending on what's wrong. Even with all this heating up, my rad STILL rapidly cools down the CPU because, within several seconds the CPU temps are already a few degrees above ambient. The Notte is powerful.
12. I record the temperatures in my spreadsheet and use the same formula for each to display the delta T. This is extremely important because when trying to rank the various configs it is impossible to do so correctly if I don't use delta T. I also record other things, like clock speed, CPU/Thread usage, CB & TS scores & FPS, and total CPU usage (average and max). This sometimes alerts me to an anomaly in testing that otherwise wouldn't be evident from just temps.
13. Again, I'm looking for formulas. Please.

 

[ShrimpBrime]

You'll make a formula that works for your personal testing.

Ambient swings shouldn't be abnormal considering you would have a thermostat on the wall that may be programmed for a temp swing. IE: furnace on 2 degree below setting and off 2 degree above setting with a total of 4 degree total swing.

Not to mention and taking into account the room is being heated by the PC which is not likely located near the household thermostat.

In my stupid much looked over opinion, any kind of testing for an AIO cooler shouldn't be necessary if the cooler was purchased knowing it can or cannot cover the wattage of the cpu, which is all it's cooling.

So no, there's not really a strict formula to give you.

Also, as a last thought, if the cpu isn't at throttle temp the entire time of usage, you'll calculate the AVERAGE wattage dissipated and come to a conclusion. Nothing more is needed for AIO testing, the average wattage of an AMD cpu 5600G will be 65w, not it's max output capable. Understanding this feature should guide you in the correct direction, the cooler is probably fine and no reason to over think it.

As always, Good Luck!

 

[bug]

In my testing, I've suspected that (because of things like resistance and heat production, captive air in a case, turbulence, etc.) when the ambient temperature changes, so will the results, but it won't be linear, and now I've proven this to such an extent that I have to find a solution, if one exists, to normalize test results. Sadly, I have a few challenges in this regard.

The formula is an open benchtable. Or, since you can't really splurge for one, just open up the case and point a fan at it. As long as you can reasonably control the room temperature, that should be enough to get reliable test results.

 

[ir_cow]

Open bench. Close the door for the room and keep a temperature panel on the table. Keep the tests within 3-5F. It's a bit harder with different seasons when working from home.

 

[ARF]

The results depend on how warm the CPU/GPU are before executing the tests themselves.
Tests right after a cold start after a several-day shut down show higher results, than the same tests performed after a whole-day use of the PC.

You need data about the temperature-based boosting algorithm and frequencies changes of your tested devices, and then add the ambient temperature as a secondary and not important variable.

 

[sneekypeet]

I believe this is where "deltas" come into play.

A few sites will not use actual thermal results due to the same issues you are running into. While it may not be as perfect as you want, it is the closest you are going to get without removing the variables.

 

[TPCEA]

@sneekypeet @ARF @ir_cow @bug @ShrimpBrime

There seems to be a bit of a misunderstanding here, so let me add some more information in my OP, and kindly refer to it in about an hour.

 

[ShrimpBrime]

@sneekypeet @ARF @ir_cow @bug @ShrimpBrime

There seems to be a bit of a misunderstanding here, so let me add some more information in my OP, and kindly refer to it in about an hour.

Most of it was about your financial status. So I skimmed a lot of the first post.

My apologies.

But do include exactly what you are testing and why with specifics.

My short question is: what's there to test using an AIO? (If you care to answer)

 

[ir_cow]

Yeah I skimmed 90% of it because it was unrelated to post title.

Since there are a lot of smart people here, I thought I'd see if anyone can offer any useful help, such as formulas for temperatures of the CPU/iGPU (DTS and Enhanced from HWiNFO64), GPU, mobo and VRM, which are what I've been tracking of late for my AIO testing.

This is the only part that is actually a question.

 

[TPCEA]

Most of it was about your financial status. So I skimmed a lot of the first post.

My apologies.

But do include exactly what you are testing and why with specifics.

My short question is: what's there to test using an AIO? (If you care to answer)

Yeah I skimmed 90% of it because it was unrelated to post title.


This is the only part that is actually a question.

If you are still uncertain after reading my edit, let me know what questions you have. The bolded part is the whole point of my post - everything else is there to avoid many of the answers I ended up getting.

 

[ShrimpBrime]

If you are still uncertain after reading my edit, let me know what questions you have. The bolded part is the whole point of my post - everything else is there to avoid many of the answers I ended up getting.

I haven't a formula unfortunately.

Temp gradient should be considered for each material that thermals pass through. Myself, generally, use BTU for a measurement. Everyone else uses wattage based on the use of the cpu, but not including thermal leaking into PCB traces or possibly other parts of the system. So the wattage you see in hwinfo64 isn't exactly what goes into the water block (for example) and the fluctuations there of.

 

[TPCEA]

I haven't a formula unfortunately.

Temp gradient should be considered for each material that thermals pass through. Myself, generally, use BTU for a measurement. Everyone else uses wattage based on the use of the cpu, but not including thermal leaking into PCB traces or possibly other parts of the system. So the wattage you see in hwinfo64 isn't exactly what goes into the water block (for example) and the fluctuations there of.

Agreed.

 

[Shrek]

Difference from ambient is not sufficient?

 

[sneekypeet]

Delta is your only solution when using variables. Sorry to be this blunt.

I say use the A/C set to lower temp that you can keep, then do testing. (I test everything at 21.5 to 22.5*C) (humidity can be a bitch too

Also, while you are in a mindset of many others, you do throw us under the bus with open air systems, but in my perspective, you are not testing the cooler, you are testing the thermal limits of whatever closed chassis you choose to use

I could go on an on, but as "reviewer" all I can say is use what you have at hand, provide all of your information,and make a solid conclusion. As time goes on, hopefully you can make efforts to start removing variables so you do not need to hunt for a magic formula that does not exist.

 

[lilhasselhoffer]

Let me short this...because a complete answer would be 10+ pages of thermo.

Most energy transfer is going to be from conduction....in this case conduction of CPU to plate, plate to fluid, fluid to metal fins, and finally metal fins to the surrounding environment. This is represented as h = q/(T1-T2) You'll note that the assumption here is then that your T1 is the fluid temperature, the q is a material constant, and your h actually changes with relation to temperatures.


How does this work in practice? Well, you have an energy transfer. Said transfer is directly proportional to the differences in temperature...which if you cannot control you have to reasonably represent as a maximum temperature and energy transfer rate. This is how you get an air cooler rated for a 125 watt TDP (125 J/s basically brings the cooler to the point where "room temperature" matches the surrounding environment and you start to brute force heat by thermal mass of the cooler itself).


There is no magical and simplistic equation to "normalize" this. It's a brute force materials and measurements situation...and even with that you've probably got to rip everything down to model actual surface area interactions. This is...not the job of somebody who didn't like highschool math....and it's where partially differentiable equations would be used to represent a stabilized state...which I personally hated when it was called mechanisms in a college program that had a 75% drop-out rate. Your mileage may vary....but putting aside 99% of your post that seems to be structured to get people to want to help you...the only help I can recommend is that you could normalize to 3 different loads and conclude that your cooler supports or does not support a given thermal load.
IE...."the [insert name] will provide adequate cooling for CPUs up to 125 watts TDP...but will not support CPUs where the TDP exceeds 200 watts." This is a way to measure the thermal run-away point instead of trying to normalize results when you lack the equipment and procedures to do the tests right. It, coincidentally, also better mirrors real usage scenarios because rarely have I seen people aim for anything but "my overclock doesn't cause a thermal throttling response."

 

[Wirko]

Difference from ambient is not sufficient?

No, because some parts of the system are regulated by temperature.

 

[TPCEA]

Difference from ambient is not sufficient?

This:

To cite one example, if my ambient is 24C, the DTS CPU core temp will be 9C higher at 28C ambient, not 4C higher.

Does that answer your question?

Delta is your only solution when using variables. Sorry to be this blunt.

I say use the A/C set to lower temp that you can keep, then do testing. (I test everything at 21.5 to 22.5*C) (humidity can be a bitch too

Also, while you are in a mindset of many others, you do throw us under the bus with open air systems, but in my perspective, you are not testing the cooler, you are testing the thermal limits of whatever closed chassis you choose to use

I could go on an on, but as "reviewer" all I can say is use what you have at hand, provide all of your information,and make a solid conclusion. As time goes on, hopefully you can make efforts to start removing variables so you do not need to hunt for a magic formula that does not exist.

Blunt is my middle name. Blunt as in straightforward, not marijuana.

I'll probably end up doing exactly that with my AC now that it's warm enough to do so. Eventually, though, I need to build a temperature-controlled chamber for testing because that will cost less and be more effective than a window AC and space heater.

Well, no matter WHAT us reviewers do, ANY solution has its weak points. Open bench and lab testing are only really valuable for comparative purposes in a highly controlled situation, which is not how the vast majority of computers are used, thus negating their value aside from being easily reproduced and verified. Test dBA in an anechoic chamber and it'll be guaranteed to sound different in various PCs - but what else can we do aside from testing inside a case and in open air? There's no way to cover every scenario no matter how we do it. Soooooo...I prefer my one limited scenario that IS real-world for a significant number of users over the alternatives. I realize that many people look down their noses at this idea but, well, so what? If my data helps a significant number of users and testing done in isolation helps virtually none, whose data is more useful?

I would say I'm testing many things because I'm not locking down all the variables (frankly, I don't know how, yet) and that is not what most users do, either. So, sure, spikes and drops will happen, and that's part of the reason one cooler test takes 1.25 or so hours. I don't let my PC sit unused, thus reducing the risk of anomalous results due to updates and such, power spikes, etc. Nothing is perfect, though, and I acknowledge that my work is not perfect. Show me someone's work that IS perfect for the majority (or all) of users and you'll have shown me god. It'll never happen. Elon Musk or some other megalomaniac would have to fund the research and hire lots of competent people and so on, and it'd still take a really long time and be outdated in a few years as soon as there are relevant innovations in whatever tech is being tested, n'est pas?

I'm unhappy that my data is skewed by ambient temperature. I'm even more unhappy that now I'm going to have to rerun numerous tests just to complete my AIO ideal configuration testing and put out my results for the public. But what can I do? I'm not rich, nor do I have a lot of space, nor am I highly trained (although I'm pretty good at research and analysis, which lets me find my mistakes and try to remedy them!). Oh well. At least I can shrug. Atlas cannot.

Let me short this...because a complete answer would be 10+ pages of thermo.

Most energy transfer is going to be from conduction....in this case conduction of CPU to plate, plate to fluid, fluid to metal fins, and finally metal fins to the surrounding environment. This is represented as h = q/(T1-T2) You'll note that the assumption here is then that your T1 is the fluid temperature, the q is a material constant, and your h actually changes with relation to temperatures.


How does this work in practice? Well, you have an energy transfer. Said transfer is directly proportional to the differences in temperature...which if you cannot control you have to reasonably represent as a maximum temperature and energy transfer rate. This is how you get an air cooler rated for a 125 watt TDP (125 J/s basically brings the cooler to the point where "room temperature" matches the surrounding environment and you start to brute force heat by thermal mass of the cooler itself).

I'm following you. I personally dislike the use of watts to represent what is actually degrees/watt or something like that, and we know that TDP is flawed even when the manufacturer doesn't lie about it.

There is no magical and simplistic equation to "normalize" this.

I'm aware of that, nor did I request such a thing.

It's a brute force materials and measurements situation...and even with that you've probably got to rip everything down to model actual surface area interactions. This is...not the job of somebody who didn't like highschool math....and it's where partially differentiable equations would be used to represent a stabilized state...which I personally hated when it was called mechanisms in a college program that had a 75% drop-out rate. Your mileage may vary....

Agreed.

but putting aside 99% of your post that seems to be structured to get people to want to help you...

Not 99%, not even 10%, but you can infer what you like - it doesn't matter in the end. I didn't lie about anything, and I made it clear why I can't just throw money at the problem by buying things and/or hiring someone to figure it out for me.

the only help I can recommend is that you could normalize to 3 different loads and conclude that your cooler supports or does not support a given thermal load.

Interesting that you should mention that because that is what I do. I use 3 different loads (P95+work+video; CB24+video; Time Spy), but I also want to run these tests at different ambients.

I run Prime 95 with Small FFTs for 1 hour, which I'm sure someone will say that it's more than long enough, so please don't. I do this to make sure that the rad is as heat-soaked as possible. To aid in this, I also watch a video and do some work at the same time. I then run Cinebench 2024 CPU Multi-Core while watching a video and, finally, I run Time Spy (with no other activities since it is sensitive). If I make a mistake, I redo all or part of the testing depending on what's wrong. Even with all this heating up, my rad STILL rapidly cools down the CPU because, within several seconds the CPU temps are already a few degrees above ambient. The Notte is powerful.

If all or part of that is not what you meant, please clarify.

IE...."the [insert name] will provide adequate cooling for CPUs up to 125 watts TDP...but will not support CPUs where the TDP exceeds 200 watts." This is a way to measure the thermal run-away point instead of trying to normalize results when you lack the equipment and procedures to do the tests right. It, coincidentally, also better mirrors real usage scenarios because rarely have I seen people aim for anything but "my overclock doesn't cause a thermal throttling response."

Great - how do I take my results and use them to equate to TDP? Better yet, let's make a better system than the current TDP - we can call it TDP-LH (in your honor, of course) that actually reflects degrees of heat produced/watt. Ah, but then we run into the reality that it's not just the CPU producing heat, and that radiated heat from everything else impacts the CPU and cooler. Still, something is better than nothing, right?

 

[ShrimpBrime]

To make it simple as possible, total wattage dissipate, just use the watt meter at the plug in the wall. The energy put in is dissipated nearly entirely thermally except the fans which convert some energy to centrifugal force. You'd want to calculate the efficiency of these fans.

So watch the wall meter for very many several days with normal usage to create an average. For the entire system.

TDP or thermal design point is to help the builder pick adequate cooling. If the cpu has max boost TDP of 235w, then you may want a cooler suffice to dissipate that wattage.

Calculate via averages. Probably be the easiest approach I think.

But an equation to have better TPD isn't really a thing because we don't design cpus around the cooler.

 

[lilhasselhoffer]

This:

Does that answer your question?

Blunt is my middle name. Blunt as in straightforward, not marijuana.

I'll probably end up doing exactly that with my AC now that it's warm enough to do so. Eventually, though, I need to build a temperature-controlled chamber for testing because that will cost less and be more effective than a window AC and space heater.

Well, no matter WHAT us reviewers do, ANY solution has its weak points. Open bench and lab testing are only really valuable for comparative purposes in a highly controlled situation, which is not how the vast majority of computers are used, thus negating their value aside from being easily reproduced and verified. Test dBA in an anechoic chamber and it'll be guaranteed to sound different in various PCs - but what else can we do aside from testing inside a case and in open air? There's no way to cover every scenario no matter how we do it. Soooooo...I prefer my one limited scenario that IS real-world for a significant number of users over the alternatives. I realize that many people look down their noses at this idea but, well, so what? If my data helps a significant number of users and testing done in isolation helps virtually none, whose data is more useful?

I would say I'm testing many things because I'm not locking down all the variables (frankly, I don't know how, yet) and that is not what most users do, either. So, sure, spikes and drops will happen, and that's part of the reason one cooler test takes 1.25 or so hours. I don't let my PC sit unused, thus reducing the risk of anomalous results due to updates and such, power spikes, etc. Nothing is perfect, though, and I acknowledge that my work is not perfect. Show me someone's work that IS perfect for the majority (or all) of users and you'll have shown me god. It'll never happen. Elon Musk or some other megalomaniac would have to fund the research and hire lots of competent people and so on, and it'd still take a really long time and be outdated in a few years as soon as there are relevant innovations in whatever tech is being tested, n'est pas?

I'm unhappy that my data is skewed by ambient temperature. I'm even more unhappy that now I'm going to have to rerun numerous tests just to complete my AIO ideal configuration testing and put out my results for the public. But what can I do? I'm not rich, nor do I have a lot of space, nor am I highly trained (although I'm pretty good at research and analysis, which lets me find my mistakes and try to remedy them!). Oh well. At least I can shrug. Atlas cannot.


I'm following you. I personally dislike the use of watts to represent what is actually degrees/watt or something like that, and we know that TDP is flawed even when the manufacturer doesn't lie about it.

I'm aware of that, nor did I request such a thing.

Agreed.

Not 99%, not even 10%, but you can infer what you like - it doesn't matter in the end. I didn't lie about anything, and I made it clear why I can't just throw money at the problem by buying things and/or hiring someone to figure it out for me.


Interesting that you should mention that because that is what I do. I use 3 different loads (P95+work+video; CB24+video; Time Spy), but I also want to run these tests at different ambients.

If all or part of that is not what you meant, please clarify.

Great - how do I take my results and use them to equate to TDP? Better yet, let's make a better system than the current TDP - we can call it TDP-LH (in your honor, of course) that actually reflects degrees of heat produced/watt. Ah, but then we run into the reality that it's not just the CPU producing heat, and that radiated heat from everything else impacts the CPU and cooler. Still, something is better than nothing, right?

You are like two steps away from understanding this...but just not quite there. Let me take another crack...and maybe we can bridge the gap.


Heat is not special. It's a measure of energy stored, as random kinetic energy. A joule is a unit of energy...and a watt is a rate of energy. IE, a certain amount of joules in a material represent it is of a given temperature...while conducting heat transfers joules at some rate from an area of greater density to lower density. The measurement of TDP as watts is not wrong....the issue is that how those watts are calculated is why AMD and Intel have different TDP issues.


Hopefully you've got the whole rate versus energy thing now...because it shouldn't be rocket science. If you cannot grasp that then this next bit is going to be a problem. One last chance to jump off the train.

Now, thermal runaway is when you've got more energy being put into the system than it can effectively transfer. In the case of a liquid cooler think about it as an input rate from the CPU of 200 Watts (Joules/s), and an output to the environment of 100 Watts. Your fluid heats rapidly, until the difference in temperature between the CPU and the fluid allows for enough flow. Thing is, this could have the CPU way hotter than the 90 C thermal management that the CPU starts to throttle at....so you've effectively forced the CPU to throttle despite maybe only having a fluid temperature of a few degrees above ambient. Companies try to prevent this with a series of fins at the water block to fluid interface (more surface area, more transfer time, more energy flow possible)...which is why if you crack open a block you'll see plenty of fins.

The more complicated thing is that you can also get a runaway if the radiator doesn't have enough surface area. Fin area may be enough to transfer, but if you're restricting enough airflow to start having the cooler increase your ambient then the same delta T will require greater heats. IE, a 20C ambient with 30 C liquid loop has a delta T equal to that of an 99C loop and 89C ambient. Push up ambient, push up delta T, change heat transfer slightly because it is a property based on atomic arrangements, and you get the heating runaway.



Mopping up a little...multiple loads is different from multiple energy transfer rates. You've falsely equated TDPs here with load TDPs...and that's basically why people don't trust the Intel/AMD numbers.

 

[Solaris17]

@tabascosauz is nicer than me, play nice.

 

[freeagent]

I just give raw numbers. My ambient, my load, and all relevant numbers showing the operating conditions the system is working in.. Its just easier for me because I suck at math. I make sure to show relevant HWiNFO stuff, a couple of times I had to break out my trusty Accutemp, it shows room temperature and humidity. I try to do my testing between heating and cooling cycles depending on the season. My setup is in the basement, so I already have an unfair advantage. Its been awhile since I have tested outside of a case, but I think I prefer in-case testing, just because I have pretty good caseflow. It could go from good to crazy, but the noise gets crazy too, and I would have to switch to crazy fans.