Monitoring, loop tuning and expansion

[evolvedpc]

I don't see much discussion of cooling system monitoring in this forum. I have a fair amount of measurement equipment in my loops, which I use to help me tune my rig. After a fair bit of research and experimentation, I've decided to kick off a discussion on useful things to measure, along with what I think they mean, in an effort to both learn something from this forum and provide guidance to others. I'm no expert, so much of this is just a theory/educated guess on my part, bourne out by what I have observed as I have built my own cooling rig. I expect that any real value to be had will come from comment and discussion by those more expert that I.

I'm going to concentrate on temperature measurements for the purposes of this post - I may post about other things another time.

First, the overall measurement of a cooling system's effectiveness has to be air/chip delta. If my GPU is hitting 60C under load, with an ambient air temp of 27.5C, then I have a chip/air delta of 32.5C. Is that good? I dunno - it depends. In my case, it isn't as good as the 24C delta that I was getting before I added that second GPU to the loop, but it's good enough, as my GPU stays well under the 90C that it was happily running at, on air.

Things become more interesting when you look at what happens in between the chip and the air. Heat has to move from the chip, via your thermal paste, to the water block and into the water. Then it moves along tubes to your radiator and into the air. We can measure how we are doing at various points in this journey by looking at two new measurements: chip/water delta and water/air delta.

The difference in temperature between your chip and your water, once equilibrium has been reached, can be taken as a measure of how effectively you are transferring heat away from your chip and into the water. There are 2 main factors that will affect this: the water block you are using and how well it is mounted. Personally, I always try to buy the best equipment I can afford to buy, so the block that I use is not something I can typically improve on - it's either as good as it gets, or I can't afford a better one. That leaves the quality of the mount. In my own experience, this can vary wildly and should be paid close attention. I have had my system spontaneously reboot every time I start furmark, because I mounted my GPU block poorly. (The Koolance GTX480 block requires thermal paste on the GPU itself and 2 different thicknesses of thermal pad for the various VRM's and RAM chips - only you have to figure out what goes where by yourself, via trial and error!)

So - a high chip/water delta can mean a poorly performing block, possibly caused by a poor mount job. It is also possible that your block might be performing poorly for other reasons, such as being clogged with algae, or dye, or having poor water flow. The main thing to note is that your block needs attention.

It should be noted that your chip/water delta will be different at idle and at load. My CPU loop has a chip/water delta of 6-7C at idle and around 27C under load. I strongly suspect it is either clogged, or poorly mounted, as my GPU loop has a chip/water delta of around 6C, at idle and 15C, under load. In fact, that is what prompted this post - after analysing the temps on my newly upgraded rig, I realised that things were less than optimal in the CPU temps front - and that it wasn't simply an overloaded radiator. (I plan to rebuild my CPU loop next weekend, once I have received a few extra parts that I want to install at the same time. I'll post the results on this thread, once I'm done.)

Water air delta - commonly referred to as delta T, or dT - measures how efficiently your loop is shedding heat into the air. This will be affected (mostly) by heat load, radiator size and fan speed. Radiators shed heat faster at a higher dT, due to the simple fact that heat transfers more quickly between two mediums, as the temperature difference becomes greater.

It should be noted that, in order to properly represent the effectiveness of your radiator, dT should be measured between the water and the air flowing *into* the radiator. If your radiator is configured with case air venting via the radiator, then the temperature of the air inside the case should be used - not the ambient room temperature. In my rig, I have 2 loops - the GPU loop has a radiator mounted on the air intake, while the CPU loop has it's radiators mounted on the exhaust vents. The case air can be a good 10C higher than ambient, under load, so my CPU loop performs worse than it otherwise might, but I'm limited by what I can fit into my case.

Skinnee Labs suggests that a dT of 10C is about normal, with 5C being "good" and 15C representing a loaded loop. If you have a high dT, you need to start considering higher fan speeds, or more radiator area. If your dT deteriorates over time, you may also want to check that your radiator hasn't become clogged. Air bubbles in your radiator can also increase dT, as any place where an air bubble is touching the (inside) surface of the radiator is a place where water is not.

So, how does all of this help? Well, I have already shown how measuring chip/water delta can help you spot a poorly performing water block, but this can also be used in loop design. I mentioned previously the Skinnee Labs lists 5, 10 and 15c as being a good guideline for good, average and loaded loops. I would apply these measurements to an alternative approach, that takes a more goal-orient look at your loop design.

Basically, I want to run my loop as hot as possible, so as to achieve the highest possible air/water delta, thereby getting the most efficiency out of my radiators. I already have as many radiators in my case as will fit and I don't want to go external, or the hi-fi quality good looks of my Lian-Li case will be marred.

I also want to fit in as much heat-producing hardware into my case as possible. The more effectively I can dissipate heat, the more GPU's I can run and the more I can overclock my chips. I don't care if my CPU runs at 30C or 50C - so long as it stays under about 65C, which is (conservatively) where an i7-930 starts to stray into the realms of "too hot for comfort". Similarly, my nVidia GeForce GTX480's will happily run at 95C, so I'm not bothered as long as they stay under that temp.

So, I take the max temp I want to run my chip at, (say 60C, for my CPU,) then subtract my chip/water delta, (27.5C, at the moment,) leaving me with a maximum comfortable water temperature of 32.5C. That's only about 4.5C above today's room temperature, what with this heatwave we've been having in the UK this summer. The air/water delta on my CPU loop is around 7 or 8C, under load and measured against case air, leaving me with a CPU temp issue. (In fact, my actual CPU temps have been straying as high as 70C under combined Furmark/Prime95 testing. No actual problems yet, but not very comforting.)

On the other hand, my GPU loop, with it's chip/water delta of 15C, could theoretically run at a 75C water temp, leaving me with a huge 40C+ air/water delta to play with. However, a further limit is imposed by my pumps. A Swiftech MCP355 (like most common pumps,) is rated to a maximum operating water temperature of 60C, thereby imposing a maximum limit on how hot I can run my loop. Leaving a margin for error of 5C, I'll settle for a 55C water temp, leaving me with 70C chip temps and a 25C air/water delta - even on the hottest summer day, here in sunny Suffolk. That's going to have my radiator working as efficiently as it's likely to get. Efficient enough, I think, to get to tri-SLI GTX480's on a 140.2 radiator with 1500RPM fans. (For quad-SLI, I suspect I'll need to sort out the CPU/water delta and re-task one of the 140.1 rads from the CPU loop.)

 

[Elfie]

I'm possibly misunderstanding something obvious here but when you say that

'Basically, I want to run my loop as hot as possible, so as to achieve the highest possible air/water delta, thereby getting the most efficiency out of my radiators...'

Do you mean you want to increase your heat load in your system purely to have it operating at maximum efficiency without then adding more rads to disperse this additional heat? Now I'm no expert but my goal is to reduce my heat load as much as I can by having sufficient rads and fans and aiming for a DeltaT of around 5ºC.

 

[Spawn-Inc]

I'm possibly misunderstanding something obvious here but when you say that

'Basically, I want to run my loop as hot as possible, so as to achieve the highest possible air/water delta, thereby getting the most efficiency out of my radiators...'

Do you mean you want to increase your heat load in your system purely to have it operating at maximum efficiency without then adding more rads to disperse this additional heat? Now I'm no expert but my goal is to reduce my heat load as much as I can by having sufficient rads and fans and aiming for a DeltaT of around 5ºC.

i skimmed through the post and got the same idea. i wouldn't go for efficiency as that would mean higher operating temps. i want mine as low as i can get them, currently 2C deltaT.

 

[m0r7if3r]

I think what he's saying is that he wants a lower cwdT( chip to water temperature delta), even if that means a higher wadT (water to air temperature delta)...essentially, greater thermal efficiency in transferring heat INTO the water. I think this is good enough to be made into a sticky, given some modification.

Something i've been considering is what could be used in the place of water to more effectively transfer heat from the block to the rad...

 

[Conumdrum]

I agree. A better block, higher flow rates, magic unobtanium liquid, pure thermal paste. With magic dust tossed in. I for one want to see it happen, will it? Someday.

If we could get Intel to get better at the heatspreaders it would make a lot of difference. Maybe Swiftech and Intel can make a CPU where the chip can float in liquid?

Time will tell as chips draw more wattage whats next on the drawing board.

Evolvedpc makes some good points.

 

[Radeon28]

I'm getting the same thoughts as well, deltas are key, and basic HT equations show us that the higher the temp of one medium, the greater the heat transfer, so long as the radiator allows it. If you were able to raise the temp of you loop, or more specifically add more heat via removing it from components, your radiators would work better through shear mass flow rate. Sounds crazy, but proven math stands behind it (however, this is all limited by the maximum HT of your rad, which is defined by ambient temps).

My cpu runs about 30C idle, and the the GPU being 37C idle, with ambient at 27C and the loop at 29C... meaning my rad is working at max (with a delta of 2C, but my blocks can't keep up with the raddage (with a delta somewhere between 3C-10C). Rad efficiency is easy, you just throw more more rads into the loop, but block delta efficiency is where it gets tricky.

To be quite honest, I'm not sure if I answered anything just now, but I'm a little to to care right now (free adult beverages at the carnival tonight)!

 

[m0r7if3r]

for those of you looking to figure out SPECIFICALLY how various things cool, this wikipedia article should answer your questions, or at least get you going in the right direction (calculus warning)

 

[evolvedpc]

I'm possibly misunderstanding something obvious here but when you say that

'Basically, I want to run my loop as hot as possible, so as to achieve the highest possible air/water delta, thereby getting the most efficiency out of my radiators...'

Do you mean you want to increase your heat load in your system purely to have it operating at maximum efficiency without then adding more rads to disperse this additional heat? Now I'm no expert but my goal is to reduce my heat load as much as I can by having sufficient rads and fans and aiming for a DeltaT of around 5ºC.

Hmmm ... I seem to have been less than clear on this point. I am mixing two concepts in my explanation, here.

The central concept is that chip temp = ambient + air/water delta + chip/water delta. You probably can't affect ambient too much, but you can lower chip temps, by either improving air/water delta (more rads/fans) or by improving chip/water delta (clean your blocks, mount them more carefully, using better thermal paste, etc). Chip/water delta is a clear indication of block/mount effectiveness.

A secondary concept is that, if radiator area is limited, you can make the most of what you have with a higher dT. Radiators shed more watts of heat at a higher air/water delta. Running them hotter allows us to shed more heat per square inch of radiator. Loops /could/ be run at far higher temps (air/water delta) than is typical, if you can get the chip/water delta as low as possible.

For me, radiator area is the most difficult thing for me to improve. I have crammed all the rads that will fit into my case, without some major surgery. So, I am keen to get my system performing as well as possible within this limit. When I say performing, I mean I want the highest framerates possible - so more GPU's are on the way, along with higher overclocks. I wouldn't be able to do this on air. For me, the point of water cooling is to allow me to build the most powerful PC than I can.

It is fairly certain that (all other things remaining the same,) an improved chip/water delta will mean more of the heat carried away from my chip and into the rad - so lower overall chip temps. Some might be happy with the lower temps. I plan to use the gain to let me cram in more hardware. Others may have different goals. They might want to reduce fan speed, for quieter operation, or ...

i skimmed through the post and got the same idea. i wouldn't go for efficiency as that would mean higher operating temps. i want mine as low as i can get them, currently 2C deltaT.

... they might want to just run with lower chip temps. Chips that run colder tend to live longer, so that is a worthy goal in itself.

To elaborate further on Spawn-Inc's comment, he could (theoretically) have a 2C dT, while still having my 27C chip/water delta. Clearly, if this were the case, he would have lower chip temps (and slightly higher water temps) by addressing the chip/water issue. (Basically, your air water delta of 2C means your rad is more than capable of doing the job; perhaps a smaller/cheaper rad would do just as well?)

For me, chip temps in the "safe" range (i.e. anything lower than stock chips on air coolers,) are all that I'm bothered with - super low temps are not a goal of mine. Others have different goals, but more attention to chip/water delta may well benefit them, too.

I think what he's saying is that he wants a lower cwdT( chip to water temperature delta), even if that means a higher wadT (water to air temperature delta)...essentially, greater thermal efficiency in transferring heat INTO the water. I think this is good enough to be made into a sticky, given some modification.

Something i've been considering is what could be used in the place of water to more effectively transfer heat from the block to the rad...

Yes, I am saying that, but what I am really saying is that analysis of my 27C chip/water delta shows that *my* CPU loop could be improved by cleaning/re-mounting the block. I also suspect that, with 2 x 140.1 rads in my CPU loop and an air/water delta of around 7C, I can probably move the second rad to my GPU loop and still have acceptable CPU temps, if the re-mounted CPU block gives me the boost I'm hoping for.

I don't think there is enough discussion in the forums about the insights that can be gained from monitoring - and what you should monitor in order to achieve those insights.

I am also saying that, (as an example of those insights,) unless super-low chip temps are your thing, you can probably get away with less radiator for a given heat load than seems to be the common wisdom. There is more headroom in dT than the 15C "loaded loop" figure that Skinnee suggests - particularly when it comes to GPU loops, where chips can afford to be much hotter. If you operate within the 60C operating water temp of most pumps, you will be hard pressed to overheat your GPU, assuming a well-mounted, unclogged block. You can also do it with a fraction of the radiator area that you would need if you were to insist on you 2C cwdT.

I plan to keep on adding GPU's to my GPU loop until my air/water delta reaches 25C and/or my water temp reaches 55C. I already have 2 GTX480's, with a 3rd on the way. with a little further tuning, I think I might even be able to stretch to a 4th. (If it melts, I'll post pics for you all to laugh at!) ;-)

 

[Elfie]

I can see your point about GPUs being able to cope with higher temperatures so an increase in DeltaT isn't so important for them. I obviously misread your first post by thinking you were saying that with the CPU and GPU(s) were in a single loop higher DeltaTs weren't that important.

I do know that when my block became clogged my water temps increase by roughly 5ºC with a similar increase in CPU temps. I now check my load temps every now and then and any increase by more than 1ºC would lead me to check my loop.

When trying to lower my CPU temps it was thanks to help from people on here I finally managed to understand that it was the ability of my block to transfer heat that was the limiting factor given I had sufficient rads and that is why I thought I was beginning to understand water cooling. I must be particularly dense when I can't understand, in a CPU loop only, why increasing water? temps increases efficiency when my CPU temps increase correspondingly .

 

[m0r7if3r]

Look at it this way, currently, the majority of the cooling load lies in having the water as cold as possible, for a lower cwdT you could operate a higher wadT and maintain the same temp. Think of it all in terms of maintaining the same temp. Watercooling parts as a whole are pretty well designed, we can get the water down as to ambient pretty much, but the low efficiency of heat transfer away from the chip means that we cannot get the chip to ambient. With 100% efficient transfer away from the chip, we could achieve this. Part of the reason this isn't happening is probably because the surface area of the chips is so small, so there isn't much surface area for the cpu block to contact. This is evidenced by the fact that gpu's see a much greater drop at load for the same radiator space, their physical chips are much larger. Does that clear it up?

 

[Elfie]

Nope, sorry. Are we talking theoretical here? That is if we had the perfect water block that doesn't actually exist that could transfer 100% of the CPU heat? I still don't see how increasing heat load and water temps which leads to higher CPU temps is somehow more efficient.

 

[m0r7if3r]

The end is theoretical...well, its all theoretical, but towards the end we enter physics world. You're not increasing heatload, you're increasing the heat you pull away from the processor, which is thus raising your wadT, but your cadT stays the same. Does that make sense? The system is the sum of its parts so

dT[sub]total[/sub]=dT[sub]chip to water[/sub] + dT[sub]water to air[/sub]

if, for the same totaldT, you you shift the ratio of dT_{chip to water}:dT_{water to air}, you need less radiator space because of newtonian cooling in that higher temperature objects move faster towards equilibrium (in this case, the object is the water in the rad). Now, this means that if we can get our dT_cw:dT_wa lower, the raddage currently employed would be able to operate more efficiently because the water would be hotter. This increased heat, however, does not mean that one is underradded, it means that the cpu block is operating at greater efficiency and thus is allowing the radiators to operate more efficiently.

better? I can do examples if need be...but it'll require breaking out calculus, and that might scare people away from the thread.

 

[Elfie]

If I don't understand it now I doubt that calculus would help

When it came to understanding the relationship between water temps and CPU temps rge did me a really simple explanation that even I could understand http://www.overclockers.com/forums/showpost.php?p=6431154&postcount=9

I need something similar.

Let's not worry about though, everyone else seems to understand the concept and it doesn't matter if I don't. If I have stuff that gets hot I'll add rads/fans until it gets cool enough to satisfy me..

 

[m0r7if3r]

That whole post still holds true, but think of it like this: you've seen how a bad cpu mount can affect temps right? Well that actually DECREASES your water temp as well (albeit very little) a good cpu mount will increase your water temp (once again, nominally). If you expand this to IHS contact with the chip and better mounts, you will be transferring more heat into your loop, and by doing this, lowering your temps (because by transferring it into the water, you're transferring it away from the cpu)

 

[evolvedpc]

Nope, sorry. Are we talking theoretical here i.e. that is if we had the perfect water block that doesn't actually exist that could transfer 100% of the CPU heat? I still don't see how increasing heat load and water temps which leads to higher CPU temps is somehow more efficient.

I think you mixing up two of the points that I'm trying to make.
I'm not trying to say the increasing heat load, or water temps, with a corresponding increase in chip temps, would make your loop more efficient. What I am saying is that:
1) Ambient + cwdT + awdT = chip temp. If you can make your cwdT smaller then, for the same ambient temp, you will get lower chip temps.

2) A higher awdT means that your radiator will shed more heat for a given airflow than it would at a lower awdT.

Let's say you cleaned and remounted your CPU block, with a resulting improvement in cwdT. Your chip temp ought to go down correspondingly. (I have a theory that your water temp will stay the same, as the chip will still be shedding the same number of watts of heat, under load - it just won't need as big of a cwdT to achieve the same heat transfer rate.)

Now, if you were happy with the chip temps you had already, then you have a little headroom to play with that you didn't have before. You could choose to take the win in temps and call it a day, or you could run your fans a little slower, (and quieter,) or you might even add another block to that loop.

The latter two options will result in your water temps going up, but if you judge it right, your chip temp won't go any higher than it was before you improved your cwdT. So, your loop is no more efficient than it was, as it is either shedding more heat, (if you added another block,) or it is shedding the same heat, but with less airflow.

Why is it shedding the same heat for less airflow? Because awdT has a direct affect on how much heat your radiator will shed for a given airflow. Heat transfers more quickly between two mediums, as the temperature difference increases.

For a practical example, let us look at my CPU loop. On a hot day, (28C ambient,) I get water temps of about 41C and chip temps of up to 68C, under Prime95 testing. This gives me a cwdT of 68-41=27C, along with a awdT of 41-28=13C. I dunno what sort of cwdT you typically get from your rig, but that cwdT looks high to me. (I'd be interested to know what cwdT others are getting, actually.) If I can reduce that to 15C, by cleaning and remounting my CPU block, I should get a corresponding 27-15=12C decrease in my chip temp (down to 56C).

That's the theory, anyway. Maybe, as m0r7if3r says, with CPU surface area being smaller than GPU surface area I'll find that I can't improve my cwdT like I'm hoping - perhaps 27C is as good as it gets? I'll find out next weekend, I guess!

To take the example a little further and add another little wrinkle to the situation, the fans on my CPU radiators blow air outward, after it has been heated by blowing into the case, over my GPU radiator. So, the case air that is blowing over my CPU rad is a good 8C higher than ambient. Effectively, the awdT of my CPU loop is closer to 5C. i.e. The difference between case air and water is a more correct representation of my CPU radiator performance.

I'm running 2 x 140.1 rads in that loop. If I can achieve the aforementioned 12C decrease in CPU temps, I might be tempted to move the second 140.1 over to my GPU loop and add another GPU. To be honest, I suspect that a 12C improvement in cwdT won't be enough to let me get away with this, but a few more tweaks may well get me close enough. (Perhaps that reserator could suck enough heat away from my mobo and RAM to lower case air temps a few degrees? Can't hurt to try that. Reversing the airflow, so the the )

This last point is intended to illustrate a practical benefit of monitoring your loop, beyond just telling you when you get too hot - which is really the point of this thread. I'm walking into my next round of mods with some concrete data that gives me at least some idea of what to expect.

It seems to me that most people just go for overkill on the rads, to be on the safe side. (e.g. Can I get away with a 120.2 for my CPU loop? Maybe, but better to get the 120.3 and make sure!) I can't add any more rads, because they won't fit, but I'd still like to make my PC faster. That means I'm forced to push things closer to the edge than I could get away with without close attention to measuring stuff. What prompted me to start this thread is a desire to see a little more discussion on what all those number mean and how you can practically apply them.

 

[Elfie]

That whole post still holds true, but think of it like this: you've seen how a bad cpu mount can affect temps right? Well that actually DECREASES your water temp as well (albeit very little) a good cpu mount will increase your water temp (once again, nominally). If you expand this to IHS contact with the chip and better mounts, you will be transferring more heat into your loop, and by doing this, lowering your temps (because by transferring it into the water, you're transferring it away from the cpu)

I understand that. I'm not quite as stupid as I thought

 

[Adragontattoo]

for those of you looking to figure out SPECIFICALLY how various things cool, this wikipedia article should answer your questions, or at least get you going in the right direction (calculus warning)

I was told there would be no math.

On Topic.

I do not monitor the water temp as it isnt hugely important to me. I watch the Q6600 temp to see what it is at and keep an eye on the loop for anything that is amiss (wrigglers, green, low level, shamu) but I don't really care that my water is 2c when the chip is 32c or whatever. My Q6600 at 3.8 and the 9800GTX+ together dont get above 34c even when I am stressing the rig pretty hard. Then again, I do have a quasi Air/Liquid cooling system going on (3 fans on the Rad, 6 on the case and all are relatively high speed.)

 

[Elfie]

evolvedpc - Thank you, I also understand that now.

As I did some testing recently I have the figures for the system in my sig to hand although my sensors aren't particular sensitive but have been 'calibrated' with a known accurate one. With GPU at idle I get:

Ambient = 23ºC
Load Water temp = 28ºC
CPU Load temp = 63ºC

This was after cleaning my loop and is roughly a degree cooler than it was.

 

[m0r7if3r]

I oversimplified the physics, it won't workout exactly like that, it will look something like this

Energy[sub]total[/sub](this is the heat generated by whatever is in the loop, lets assume just the cpu)
-
Dissipation[sub]1[/sub] + Dissipation[sub]2[/sub] (each dissipation is for a radiator, it will be a newtonian cooling equation for each one OR you can be careful with your numbers and sum the radiator space...moving on)
+
Innefficiencies (whatever amount of energy is lost due to inefficiency in the system, ihs contact, thermal paste, quality of mount, effectiveness of energy transfer, etc. All of this is stuff that's above my level of physics understanding, I don't know how it behaves, probably some exponential equation...but that's just a guess)
=
0

So, its really not as straighforward as add the temps, you have to think about it in terms of energy, not in terms of temperature differences...rad dissipation increases at a given heatload until the = 0 is met, until then it will equal the energy that is being left in the loop...hope that helps...

 

[Elfie]

I oversimplified the physics, it won't workout exactly like that, it will look something like this

Energy[sub]total[/sub](this is the heat generated by whatever is in the loop, lets assume just the cpu)
-
Dissipation[sub]1[/sub] + Dissipation[sub]2[/sub] (each dissipation is for a radiator, it will be a newtonian cooling equation for each one OR you can be careful with your numbers and sum the radiator space...moving on)
+
Innefficiencies (whatever amount of energy is lost due to inefficiency in the system, ihs contact, thermal paste, quality of mount, effectiveness of energy transfer, etc. All of this is stuff that's above my level of physics understanding, I don't know how it behaves, probably some exponential equation...but that's just a guess)
=
0

So, its really not as straighforward as add the temps, you have to think about it in terms of energy, not in terms of temperature differences...rad dissipation increases at a given heatload until the = 0 is met, until then it will equal the energy that is being left in the loop...hope that helps...

Not in the slightest but thanks for trying, it might help others.