Can someone with some real knowlege answer this question???
 

Here is the million dollar question that is driving me crazy with all this water block contruction talk. I hear people say blocks are reaching there limits and I here from other they have a long way to go to reach there limits. So here is the question: WHAT IS THE LIMIT????

There has to be a spot where everything is as optimal as it can get and there will be a certain temp between the water temp, air temp, and CPU temp. This number should remain constant through various wattaged ext..

Say your air temp is 25C, water temp is 30C. It is impossible for the CPU to be 30C so what is the lowest possible temp that can be attained with a Copper and Aluminum block. There has to be a spot that it can get no cooler. (this is just an example)

This should be a mathmatical answer that requires no testing. If we can come up with this number then we know when we are getting close to the limit and we know when to stop coming up with new designs.

Am I off base? Am I wrong to think this number should be the FIRST thing a block designer should come up with before he/she starts building performance blocks? Am I wrong to think that there is a mathmatical answer to this question. And if there is a mathmatical answer who can come up with it?

This answer would sure answer some critical questions.

I would think you should be able to take the thermal properties of Copper and Water and figure up an answer. There has to be a point where it just can't get any better.

I'll take a stab at this daunting question...

The point at which it gets no better (unless we start talking about the water evaporating and chilling as it moves through the air) is the CPU at the same temperature as the water.

The problem here is practical limitations to what you can achieve.

The theoretical best you could do is direct die cooling with a perfectly formed impingement jet of water covering the entire die surface. Now as the velocity of the water is increased, so is the convectional coefficient of the water striking the die surface. Is there is limit to how high it can go? No, not really. Practically? Yes. Can't inject the water at the die surface at 1000m/s and drill a hole into the die or wear it away in a few seconds, or crush it under the sheer force of the water striking it. Not mention what do we now do with all this water. How to we route it away, etc.

Okay, so let's not blast the die directly and have a water-block instead so we can route the water away and not damage the CPU. What works best here is an infinitessimally thin base-plate. Not practical again. 0.1mm thick? Also not practical. How thick then? That depends on the mounting pressure and the internal design of the block and how it braces the base-plate. Okay, so we build our block, now how fast can we blast the water at the block and not wear away the copper?

Basically what I'm saying is that there is no way to easily answer your question. It can be answered given a very particular set of parameters and substances, but also it's a case of how long do you want it to last? How robust is it? etc.

What I'm saying there's the theoretical best, which is easy to answer. CPU die = water-temp, and then there's the practical best, but this becomes suddenly subjective on your definition of practicality (10hp positive displacement pump? cost? etc).

What's the practical limit then? Well that depends on your budget...

Then wouldn't one of those little copper blocks that a basically a square on the outside tube that water runs though, be the best block? With a good high flow rad and fast pump?

Cathar is singling out or at least emphasizing flow and thickness of the copper, there's more to it than that of course including surface area and temperature gradient. Although desings like microchannels, Maze, or Spirals chief purpose is surface area, they can also increase turbulence, which increases thermal convection at the thin layer of water touching the copper.

It's not all about the block though is it?, it's about lowering that 30C water while still in a 25C ambient as well...

It's a damn good question and I'm sure there IS an answer, meaning yes there is an optimum available transfer for Cu & H20. but I fear it'll be only relevant to a specific scenario, not transferable across different temps and heatloads :( ... and it still won't tell us how to achieve the best liason, there'll always be block design.

I hope someone scientific/definative can answer this with authority. it's too mind bending for me :drool: :) ...

Well something we'll need is cpu's with in die thermal sensors that are actually designed to be accurate, rather than "good enough to prevent thermal melt down and not much else", like we have now...

it is something bothering me too... and I have been at it, trying to find that limit...

I however do not listen to anybody telling my it's impossible to do anything... such people belong in the government, and they should stay there.... as far as I'm concerned anything and everything is possible... :)

So far, as reputable as my methods of measuring is, or not,...... the closest to that limit I came is what I'm running at right now....

air:23.2ºC
water 28.6ºC
CPU0: 33.1ºC (Duron 1g at 1150)
CPU2: 33.0ºC (Duron 1g at 1150)
CPU3: 36.1ºC (athlon 1700+ at 2000+)

I guess it's like doing the 100m sprint.... as long as there is a difference between water temp and core temp, there is room for improvement...

the question:

"There has to be a spot where everything is as optimal as it can get and there will be a certain temp between the water temp, air temp, and CPU temp. This number should remain constant through various wattaged ext.. "

now we are going to focus on the word optimal, eh ?

while it is tempting to think/believe/hope that there exists a single 'best' solution, such will be ever elusive
and the reason is simple; given that CPU cooling is the end effect of the interrelationship of a number of factors, then it is clear that different 'manipulations' of those factors may yield the same (or effectively indistinguishable) results

those not having followed Cathar's quest (20+ pg thread on OCAU) are encouraged to read it here
as most of the variables are discussed (or perhaps cryptically alluded to)

consider Intel's general description of CPU cooling: a cascade of gradients

if one focuses VERY hard on EACH of those individual gradients, the means to better cooling solutions become clear
as do the very real limits

to look to the wb alone for better CPU cooling is to see less than half the 'problem'
what about that TIM joint ?
and worse yet, what about the impending arrival of IHSs for AMD CPUs also ?

a cascade of gradients, indeed

returning to the wb (only, and it is less than half the total thermal impedance):
in an absolute sense it can be viewed as a coolant delivery problem, as #Rotor has understood for some time
but few are willing to address the problem in only those terms because of the cost
so now we begin to wander through a literally endless series of combinations and permutations of compromises all related to economics
-> but we do not describe our activities that way, in an honest fashion

we gussy-up our economic compromises with 'technology' bs, thereby really confusing the issue

I am working on several 'high tech' approaches quite different from conventional CPU wbs, but they are also 'high dollar'

an additional note re the originally phrased question:
all my testing to date demonstrates that there is NO change in a wb's performance due to changes in the applied power - period
(as long as the die size is the same of course)

$Rotor
go back and read how I defined T/W, using a RTD or TC embedded in the wb bp
you otherwise include the TIM joint which is an offset - and a variable

We could approach the question from the "water" angle: How much heat can water take away?

The CPU emits a varying amount of heat. Optimally, all of that heat would be taken away, but of course it's not feasible, or at least not practical to do it that way. Imagine the required size of the radiator to cool the water back down to ambient!

The limit that we're trying to break though, is how much heat energy we can extract from the CPU core. This must be the most significant barrier to the cooling solution, and I come to that conclusion from seeing many people having varying results, strictly from improving the TIM joint.

Once a good TIM joint is made, the next step is to optimize the heat transfer to the fluid. Fins are the obvious answer, turbulent flow is the necessarily added factor. The combination of both should get us pretty close, as Cathar has demonstrated.

I think the real question here is: how come there's no single answer? How come there is not a single design that is standard, and technically proven to be optimal, beyond all others?

Part of the answer lies in the heat source, and its spread pattern, IMO.

As others have already stated, the theoretical limit is basically zero delta-T. Practically speaking it's another matter. Here's an analogy for the situation.

Say you need to walk 60" (roughly the maximum delta-T from die to air in °C). Say with each step, you get to go 1/3 the distance that remains. Each step costs you $40.

For your first $40, you get to go 1/3 of the way, or 20" (40°C delta-T from chip to air, roughly 60°C true chip temperature). This is your basic air-cooled heat sink.

For another $40 ($80 total), you get to go 1/3 of what remains, or 13.33" (26.67°C delta-T). This is a top-notch heat-sink/fan combination.

For another $40 ($120 total), you get to go 1/3 of what remains, or 8.89" (17.78°C delta-T). This is where basic water cooling can get you.

For another $40 ($160 total), you get to go 1/3 of what remains, or 5.93" (11.85°C delta-T).

For another $40 ($200 total), you get to go 1/3 of what remains, or 3.95" (7.9°C delta-T).

For another $40 ($240 total), you get to go 1/3 of what remains, or 2.63" (5.27°C delta-T).

OK, so not the best analogy. The point is that it is easy to make progress when you've got a lot of delta-T laying around. As you get less and less differential between the CPU and your cooling medium it takes more and more effort to make additional gains.

It's a lot like efficiency in energy converting operations. We can burn gasoline in our engines, but only about 30-40% of the available energy is used to propel us down the road. The rest is lost as thermal energy, drag, etc. Manufacturers spend literally billions of dollars to improve engine technology, reduce drag, etc., and our vehicles get better mileage than thirty years ago, but are still very inefficient. Some concept vehicles may push efficiency over 50%, but cost millions to develop and would cost $100,000+ to produce (all while offering inferior "performance" in acceleration, handling, etc.).

Some times that which sounds so simple is really quite difficult.

The really funny thing is that someday someone may come up with a totally new approach that greatly improves the situation and we'll wonder, "why didn't someone do that before?" How about a high-efficiency peltier with a thermally-regulated supply voltage? Hey, if someone could do that, sign me up! Someday someone probably will, but that day hasn't come, yet.

no Ben

all of the heat IS taken away (if equilibrium is attained)

you have forgotten all that has been discussed about gradients, what they are, why they exist

search here, search google
but understand gradients - or you're drifting in the flow

or read - with deliberation and understanding - Cathar's thread
its in there too

and as usual you're right:mad:

Ok, let me try again, with the gradient in mind...

CPU heats up to 300+deg C, with no cooling whatsoever (and burns up). That's no (zero) steps, in myv65's analogy;) .

If we add a heat sink (no fan), the purpose of which is to spread the heat then release it, CPU temps climb very high (100?), and the CPU burns up.

If we put a fan on top of the heatsink, the cpu will remain operational. Temps range from 45 to 60 deg C (roughly, depends on CPU).

If we replace the coolant (air) with a denser one (water), the temps drop to a range of 30 to 40 (roughly).

The purpose of all this is to demonstrate that the internal temperature will vary, according to the cooling solution. The heat generated is completely dissipated, in every case, where the balance point has been achieved, i.e. the heat generated is equal to the heat dissipated.

If we supercool the coolant, temps can hit freezing, but that falls outside of the original question: what's the best we can achieve with Alu/copper?

The best what? Best temp?

The closer we get to the heat source, the lower the potential temp can be achieved.

Sidenote: ok, so if we're getting closer to the heatsource, and we're still dissipating the same amount of heat energy, why is the core temp going down? Are we dissipating more energy? No. We're taking away the buffer zone, the heatspreader, which is allowed to retain a certain energy level, as it's transfering heat. This energy level stored, is what dictates the resulting temp. Given that Alu will store less energy, wouldn't it be best? no, because it can't dissipate the heat well enough.

The problem is this: the heat source is within the core, and not at the surface of the core. What that translates into, is that there will always be a buffer zone between the actual heat source, and the coolling solution. In other words, it's not possible to cool down the core to ambient (at least not without supercooling).

So what's the best temp we can achieve?

The answer is: what's the best temp that can be achieved with direct die cooling?

BillA: we need another one of your c/w graphs, at different flow rates, for direct die cooling, with a cooling solution normalized to 20 deg C.

direct die is not an option with current CPU packaging
ergo a non-issue

Ben
you're still missing the gradient 'visualization' (and I'm inept at dwgs)
[read more - post less ?]

graph, with a pencil. the guesstimated temps at the various 'points':
(no point in useless speculation about the 'internal' CPU temp)
CPU face/TIM joint
TIM joint/wb bp face
internal bp face/coolant (temp AT the bp face proper)
coolant temp AT the bp/avg coolant outlet temp
avg coolant outlet temp/avg coolant inlet temp

connect the dots
each of those lines represents a gradient within the CPU WCing 'system'

now address each of those, serially at first, then also as they interrelate with the adjacent gradients
- some are fairly straightforward, some hideously complex

it is this 'process' that Cathar was able to understand, and address
-> with apparently excellent results

Ok, I see your point.

Core to TIM, TIM to baseplate, baseplate to coolant.

I'll give it a shot. Let's say we have a coolant temp of 30deg C, and a core temp of 40.

Core to TIM: that falls in the "complex" for me, because I don't know the thermal properties of the core material, nor how the surface is finished. If the core temp reads 40, I'm going to guess that the core surface is 42 (whoever said that the internal diode gives an accurate reading?). At the TIM, I'd guess 42, since it doesn't dissipate any heat, except to the waterblock.

TIM to bp: from 42 at the TIM, the baseplate won't be very far either. depending wether the temp is measured at the bottom of the bp, the temp should be closest to the hottest point (41.9). (a tiny bit of heat dissipated to air?)

bp to coolant: The copper and coolant should be different, as there would otherwise not be any heat transfer to the coolant. It doesn't seem to be difficult to get the copper hot, but much harder to get the heat into the coolant. In other words, the most amount of resistance in heat transfer, is in the copper to coolant transfer.

(ok, now I have to try mercury cooling!)

dig deeper Ben

some of these can be calculated
what will be the temp offset from one side to the other of C110 Cu with a hot side temp of 40°C and a thickness of:
0.075" ?
0.150" ?
0.300" ?

(there are wbs with these actual thicknesses)
assume bp same size as die, no edge effects, uniform heat flux
- finished ?

now go to waterloo and play with some more realistic assumptions

go to the AS site and see how Nevin looks at the issue
(though I have difficulties with some of it)

an aside: you cannot start each thread as if it were a brand new world, at some point you have to start incorporating what has gone before into your 'body of knowledge'; this is basic matls sci stuff - on google, all over
and the same applies to the thermo stuff, we're tilling the same field over, and over, and over; identify the principles involved - and apply them to each 'new' issue

and I repeat, read Cathar's thread; and re-read it until you understand the PROCESS by which he made choices

Jeez, isn't it more practical to either call it quits when reaching 40 degrees, or to just buy a faster processor? :D

It would be cool (so to speak) to look at some watercooling solutions of companies who really have some money and brain power to throw at it. What does the big server market say on the subject?

The big server market uses a well-worn maxim in the market: good enough. The economics of the situation dictate that R&D must be stopped when the returns do not outweigh the investment.

Big iron in the 70's used watercooling, and it was generally despised because of the complexity and (therefore) low MTBF. So, the big server market uses air cooling now. It is much more reliable and it is also cheaper to produce. Chips are fast enough for the server market such that the decreased reliability and increased manufacturing costs do not justify watercooling.

Now, if you want to throw out cost, you end up in situations where you use phase change cooling and live with a potentially inferior junction. It is cheaper to throw another stage on the phase change cascade than it is to perfect the heat exchanger. Low production volume. If you are interested in a chiller manufacturer, look at Temptronics. They produce chillers for the semiconductor test environment. They use phase change to cool a liquid and then route the liquid to the heat exchanger on the backside of the wafer.

Yeah, tell that to the guy trying to figure out how to tweak a few more hp from a stock engine. Not disagreeing, just making a point.

The big server market says KISS. You won't find cooling gear of the sort required for outrageous overclocking if you poke around a server business. They go for effective, reliable solutions, but not ones that push the envelope (unless the envelope IS reliability).

yup,
just buy a faster processor

as stated, it is the reliability aspect that will keep WCing on the (lunatic) fringe

can one imagine Yahoo or eBay with their 10s of thousands of CPUs chugging along using aquarium pumps to cool their processors

lol

I have fun testing, and others WCing and OCing - but its 'for sport', for sure

most of Cray is LC, has been for decades...

http://www.cray.com/products/systems/t3e/index.html

mid way down


Liquid cooling has been around much longer than we might be aware of...

I know, I know, perhaps I shouldn't say anything, having two waterblocks on the ready for a dual Athlon system... purely for home use... :rolleyes:

Roto# I dont disagree but comparing a CRAY, or IBM Mainframe which were also cooled with chilled Flourinert is 100% different than our rigs.

The machines above use an endless supply of chilled service water ( normally chilled as part of the HVAC system for the building its in). From there the transfer from the chilled water to the Flourinert happens over another heat exchanger. Then from there multiple pumps (with redundancy) move the coolant at high pressures and velocity through the machine. One pump for an IBM mainframe costs around 15,000$. Some of the old Crays also used LN2 to chill Flourinert and cool the system.. .the components for that cooling system is in the millions of dollars.

See Air cooling has none of those expensive requirements and risk that the LC setup has used. Today can provide CLOSE to the cooling that many of the consumer watercooling systems can.

Using watercooling on any production machine ( using consumer products) is insane I think (any machine which you rely on and depend on). that is EXACTLY the main reason I dont run watercooling in my main PC at home (while I do have it in a couple other machines)... I needed one machine that I will not need to worry about die'n due to a small leak or pump failure, or anything. Besides with todays HSF's I wouldnt gain THAT much performance ( just maybe slightly quieter.)

Its all about MTBF as bri said. The only way a cooling system becomes feasable in the top end super computers is when its built with parts and engineering that cost millions to develop. or they just use HSF's and some fans which give the same MTBF but just a lil louder :)

The fast athlon Linux cluster is nifty and all, but I see the watercooling aspect as being quite a big investment for a small payout over what you would see with a modified air cooled setup. Also for the cost of a big rig like that... Prolly could have bought faster CPU's :p

Also

Heatpipes. I recently got 2 new Dell PowerEdge 2600's in, I opened them up and under the air tunnel that they cap the CPU's with, the HSF's are about 7" tall and use 3 Heat pipes to hold about 20 aluminum plates.

with a Xeon 2.4Ghz running at full load the HSF never got more than just very slightly warm with 2 92mm fans pulling at just a hair over 1500RPM... I seriously doubt going to watercooling would help that thing at all since it already is doing such a great job with the HSF.

BTW I am talking with some Dell folks about getting a couple of those HSF's for myself.

Dang, is that where we are?

Ok, I copy-pasted it in Word, so I could print it nicely. It's not 20 pages, it's more like 160! I got it down to 79. I'll re-read it this weekend.

(I didn't get anything from the first half, except for a caricature of you:http://www.employees.org/~slf/concept/billa.jpg ).

Have a nice one!

Might I had to joe post that most big iron server are in server rooms, where the climatisation system is more noisy than the server itself....
(dell from 1650 to 6550, compaq san, storatek wolfcreek, ibm p390, ibm as400 of many size, sun enterpise 220r to enterprise 6500, they were all less noisy than the climastisation system, even the 6500 that was in another room since the climatisation system was from the floor like most server room)
So even if they cool there server with redundant noisy fan, it doesn't matter at all since no one is in the server room to ear it.
I had a Dell 2450 and 2550 rack mount 2u server with 1 p3 on the first and 2 p3 tualatin on the second. They were noisy, really noisy, but still we were using them in the server room because noise isn't really a factor there.