Some Interesting/Confusing Data from testing

[pHaestus]

It was initially my goal to estimate CPU wattage by using the following formula:

(Flow Rate*delta T (waterblock outlet - water inlet) *4186)/60

With flow rate in lpm and water temperatures in C

Presumably this should result in a constant value for W over flow rate tests as CPU MHz and Voltage are unchanged. What you actually get though looks like this:



Good agreement from block to block for the temperature rise across the wb; this is encouraging. Some difference doesn't surprise me much because the 3 barb blocks are harder to insulate than the 2 barb blocks. Look what happens when you use the above equation to calculate W:



Les and I have talked about this a bit in the past and are operating under the theory that as flow rate increases, C/W goes down, and wb efficiency (the percentage of the CPUs heat that goes through primary cooling path) goes up.

Anyway the big shift in W estimated by the equation is the reason that I haven't been using C/W for wb tests. Any advice on how to test/think about/explain these results would be most appreciated.

And for the calculation-minded, here is a spreadsheet

Caveat: The DeltaT (CPU-wb in) from these tests were adjusted down to the average value at 1.50GPM for the review I just published.

[jaydee]

I think your problem is secondary heat loss through the bottom of the CPU, the socket itself, and the pins. The more efficient the block is the less the above secondary losses will happen because there is less heat to escape. The worst the block performs the more heat to go out the secondary paths. I think it would be near impossible to get a solid CW without completely insulating the Socket. Which is probably impossible. You maybe surprised (or maybe not) how much heat flows through the pins and escapes through the traces and the PCB.

Also another notion, if the CPU is cooler it generally runs more efficiently and thus uses less wattage to achieve the same goal. [EDIT] Hummm, I might off on this on second thought.[/EDIT]

A combination of the above is possibly the problem.

Thats the best I can come up with. I can be way off base but it sounds good to me right at this minute.

[rocketmanx]

Looking at C/W plots for various water blocks it appears to me that the blocks efficienecy doesn't go up linearly. As flow rate increases C/W looks to level off.
At the same time the delta T of the air and various block parts that jaydee mentioned is going up. So the parts that are losing heat are adding more and more to the loss.
This would seem to make sense looking at the RBX. It has the most surface area of all the blocks where heat would be lost.

Seems, like jaydee said, a compound problem.
Who knows though I may be reading you're chart all wrong

[UNDERBYTE]

1.) Water in is increasing in temp so the radiator is not dissapating the heat eff. at increased flow?

2.) As to the secondary heat paths - cpu is around 30 C. what is the motherboard temp from local components 50-55C?

if anything the heat flow from secondary sources would be from + to - to the block rather than from..??

Just a thought

[jaydee]

Originally Posted by UNDERBYTE

1.) Water in is increasing in temp so the radiator is not dissapating the heat eff. at increased flow?

2.) As to the secondary heat paths - cpu is around 30 C. what is the motherboard temp from local components 50-55C?

if anything the heat flow from secondary sources would be from + to - to the block rather than from..??

Just a thought

Your #1 is irrelevant as he uses a chiller to control the water temp. That is why I think it is in the secondary heat losses. And as rocketman pointed out the mass of the block may add an extra level of secondary heat loss confusion.

Have to ponder you #2 point another time. I am way to tired at the moment to think.

[UNDERBYTE]

Point # 2 is not complicated - all heat is on a gradient if the block / CPU is cooler than the Local components / mother board , the heat is only going to go one way.

Most thermal analysis alla` intel assume that the air cooled CPU is at 60C and the board at 50C hence secondary heat paths.

you water cooled guys are way cool - different ball game it is just the opposite

[jaydee]

Originally Posted by UNDERBYTE

Point # 2 is not complicated - all heat is on a gradient if the block / CPU is cooler than the Local components / mother board , the heat is only going to go one way.

Most thermal analysis alla` intel assume that the air cooled CPU is at 60C and the board at 50C hence secondary heat paths.

you water cooled guys are way cool - different ball game it is just the opposite

Yeah I am just having a hard time adding that into the C/W confusion. The components, I wouldn't think, would run much cooler or warmer per block being the load is somewhat consistant. Seems it would zero itself out. Unless if the CPU was running cooler and more efficent it might make those parts run cooler aswell and add less heat. Thats interesting. If the block has a bigger mass or is wider, it might pick up more heat from those parts. Like the RBX though that is narrower it may be shielded more from those hot parts....

If this is the case then changing mobos out may contaminate the tests being those parts are closer or farther apart depending on the board.... Not what pH wants to hear.

BAH, to much thinking. I am beyond tired. It was a long day at work today...

[Les]

Morning and thanks for Excel..
Started to play thinking on lines of 2ndary paths and dP*Q - ala Tecumseh.
Dunno but initially trying 1/dTwb v Q plot
Unfortunately my Excel97 seems to play weird games with "DTek and Maze4" data in your Excel(2000?) - copy and Paste Special. Will take a little time to sort.

My first impressions of results "Oh God. Why did I ask".Think "Confusing data" may be an understatement.

[Les]

Spent a morning playing
Am no wiser than at end of Incoherent's thread
Dunno from where the Watts are coming
Maybe Incoherent's technique would reveal if caused by measurement errors.
Will have another look at his data. However the number of variables makes analysis a nightmare.
Possibly he could be persuaded to do some further work with fewer variables.

EDIT. The mandatory graph:

Have used "CPU diode raw" to calculate "C/W" in this instance. "CPU diode raw" is the diode temp associated with the dTwb measurement?

[UNDERBYTE]

Originally Posted by jaydee116

if the CPU was running cooler and more efficent it might make those parts run cooler aswell and add less heat. Thats interesting. If the block has a bigger mass or is wider, it might pick up more heat from those parts. Like the RBX though that is narrower it may be shielded more from those hot parts....

If this is the case then changing mobos out may contaminate the tests being those parts are closer or farther apart depending on the board.... Not what pH wants to hear.

BAH, to much thinking. I am beyond tired. It was a long day at work today...

The chiller is a reliable constant and so is the internal diode. This just leaves the MOBO and blocks as variables

In electronics cooling mag (JEDDC)precision testing usually involves encapsulating device front and back (circut card or mobo) with a water cooled setup to filter out the thermal noise.

In another set up they use copper wings/heat sinks attached to the mobo/card and set up on an islolated stand in a wind tunnel to cool the board to ambient for the same purpose as the water to filter thermal noise.

For simplicity's sakeI am an advocate of the bill adams test block - Watts in/ watts out with good degree of Accuracy. and you are not trying to extrapolate what all the test data means. Then use that as a baseline for "real world testing" and see if results hold up

After All what you really want to know is which configuration is better or worse right? Test block will tell you that. Otherwise unless test setup is more refined the best guess is just that, a guess.

[KnightElite]

Actually, there is no chiller at the moment. Just a radiator.

[BillA]

do not be mislead by data not understood

the thermal resistance of a device is independent of the applied load
period - full stop
(except I am now testing a device which is NOT !?!)
EDIT: and will, of course, decrease with an increase in the flow rate (sorry Les)

with sufficient accuracy one will 'see' some non-linearity, but this is related also to secondary losses
- large heat dies are 'bad' due to the increased difficulty of insulating them (including the upper surface)
- and CPU heat sources are vastly more difficult (try placing some TCs in the gap behind the mobo, then ck the uniformity of these temps between one test series and another)

"Les and I have talked about this a bit in the past and are operating under the theory that as flow rate increases, C/W goes down, and wb efficiency (the percentage of the CPUs heat that goes through primary cooling path) goes up."

EDIT: deleted a bunch of confused junk

so what is the latest on the chiller saga ?

[Les]

Originally Posted by unregistered

"Les and I have talked about this a bit in the past and are operating under the theory that as flow rate increases, C/W goes down, and wb efficiency (the percentage of the CPUs heat that goes through primary cooling path) goes up."

the above is ok if you delete "C/W goes down", but is an apparent increase of ~50% appropriate over the flow range tested ?
seems a bit much, no ?

See no reason to delete "C/W goes down".
Otherwise, as indicated previously , agree

[jaydee]

Originally Posted by unregistered

so what is the latest on the chiller saga ?

I thought he was using a chiller. Guess I will re-read the article again... :shrug:

[pHaestus]

re: Chiller:

Still completely up in the air. UPS will only deal with the shipper, who in this case is considered to be the Hamilton ON Mailboxes Etc.

[jaydee]

Originally Posted by pHaestus

re: Chiller:

Still completely up in the air. UPS will only deal with the shipper, who in this case is considered to be the Hamilton ON Mailboxes Etc.

Doh!


Another thought, you did changve mobo's in the middle of testing these blocks right? Would it be wrong to think a different mobo has different voltage regulation and may add or subtract a few volts changing the watt output? I know every different mobo I have reads a different voltage even if i use the same power supply on all of them.

[pHaestus]

Changes in voltage should translate to changes in the DeltaT (CPU-water in). No statistically significant change was noted (I changed after running 5 reps of the DTek Whitewater). I also reran the RBX and the performance fell right at the average value of the earlier run. MBM reports same idle and load voltage as well.

This was something I was concerned with though; that's why I reran those blocks

[jaydee]

Originally Posted by pHaestus

Changes in voltage should translate to changes in the DeltaT (CPU-water in). No statistically significant change was noted (I changed after running 5 reps of the DTek Whitewater). I also reran the RBX and the performance fell right at the average value of the earlier run. MBM reports same idle and load voltage as well.

This was something I was concerned with though; that's why I reran those blocks

And with that I will have to bow out and sit on the sidelines. If it isn't something mentioned above I just don't know. :shrug: Over my head....

[bigben2k]

Assuming that the thermal interface has a fixed C/W, the dT readings wouldn't be affected by it, under the same power.

Given that the readings indicate a max power to be ~110 W, is the CPU in a position to give out such power?

[pHaestus]

Well...

The delta T seems reliable enough (the wb in and out probes report the same number when placed in water together) and the flow rate is also imo pretty reliable (I installed the swissflow flowmeter in that loop "just to see" and it was within 0.01GPM of the GPI). The equation is good, so that wattage MUST be put into the water by the CPU. But how much of that observed W is coming from northbridge/ram/mobo power elements through traces and up into CPU?

No ideas here; need perhaps W estimates vs coolant temperature?

[KnightElite]

I doubt you've got much additional heat coming into the CPU from the other components... as you know heat propagates from warmer areas to cooler areas. It seems much more likely that as the flow increases, and the waterblock cools better, secondary cooling paths (through the pins and the motherboard, etc...) become less important, because of the higher delta-T between the block and the CPU die.

Also, all the other components should be generating a constant quantity of heat, given that their cooling is not changing at all during the entire process.

[prandtl]

Originally Posted by KnightElite

I doubt you've got much additional heat coming into the CPU from the other components... as you know heat propagates from warmer areas to cooler areas.

I remember I read somewhere (cant remember if it is Anandtech.com or hardware.fr) that the mobo voltage regulators can get very hot (in the 90ish C) and pHaetus already stated that the nb was passively cooled and was running hot too (hotter than the cpu to my understanding). As you say, heat goes from warmer to cooler... ie (nb, volt reg) - Q -> (cpu)

[freeloadingbum]

I would have expected cpu wattage to be fairly consistant with cpu temp across all the blocks but that is not the case. The maze 4 is over 35 watts higher than the cascade at near the same ( the maze is still warmer) cpu temp, the maze 4 at 2.5gpm compared to the cascade at .5gpm. Why is that? I would try retesting without the insulation and see if that's the problem.

I would consider swaping the sensors with each other and see if you get the same results. You could also try running the test at all the flow rates with the computer off and see how the sensors are affected. I would think at some point you would start to see some heat from flow friction within high pressure blocks like the cascade.

Another possible problem may be a difference in temperature gradient across the flow through the half inch tube. High pressure blocks may mix the water better than low pressure blocks giving the output sensor different readings for the same wattage (though I think this is extremely unlikely).

In the end you may never be able to separate cpu wattage from secondary losses or gains but I think it would be interesting to find out why blocks vary so much at the same cpu temp.

[lolito_fr]

I was also thinking of heat produced in the block due to friction, and conversion of potential energy into heat. For the Cascade, 2.5m @ 7.5lpm turns out to be roughly 3W unless I'm missing anything? Not entirely negligeable, but not even close to 25W. Plus this would be a quadratic variation, rather than the quasi-linear one that we're observing.

I find it hard to believe that secondary losses could amount to 25% variation in total power dissipation, considering that the sink temperature is apparently not varying by more than 0.5°C, and even the core is only changing by 2°C. (Cascade data)
Nothing can be that closely coupled to the waterblock for such a small change in temperature to produce such a large increase in power?

Non uniform mixing of flow, or sensor temps influenced by pipe surface temps seems like a good possibility.
Also PH, what are the specs for the flow meter?
(How much do you trust it…?)

[freeloadingbum]

For the Cascade, 2.5m @ 7.5lpm turns out to be roughly 3W

I've googled for hours trying to find a formula to calculate wattage based on pressure and flow but came up empty handed. Could you share yours with me? Thanks in advance.

I wonder how much of a performance enhancement you would get from mounting a fan in back of the motherboard, considering how much of an issue secondary heat seems to be.