Today's lesson: Thermodynamics 101 (crossposted in Cooling)

Or: How To PROPERLY Cool A Computer in 20 Easy Steps pics

(yes, I crossposted it, I thought it belonged in both forums)

So, for today's lesson, we will learn the basics of cooling a computer.

Simply put: cool air in, hot air out. Since the first law of thermodynamics states that temperature "flows" from hot to cold, this is critical to keeping our rigs cool. Especially in these times, when you can fry an egg on our CPUs and GPUs.

So, what's one to do?? In my case, get a good case(hurr), and then do things to it to improve the flow of cool air in and hot air out.

To wit, my riglet:


As you can see it's a nice CoolerMaster 640(iirc), with some nice features such as open flow, big 120mm variable speed fans, and plenty of places to put said fans.

Airflow:

Pretty good, if you ask me. There's a fan behind the processor as well, nice feature of this case. Also, the white "X" means I closed that particular vent, because with it open it just pulls cool air in and right out, instead of forcing the flow to go over components.

Now, the best way to air cool a processor is to get the coolest air possible to the HSF; and the best way is to run a vent from the HSF right to the outside of the case.

Which I did:




Yep, some 6" dryer duct, a 4" PVC toilet flange, and some black spraypaint.


The rig as it sits before the mod:

You can see the side fan that's over the vidcard and "southbridge", and the CPU HSF thru the side panel vents:


Closer look of the HSF:


Mmmmm, i7-920 goodness, despite the stock HSF:





You should be able to figure what's coming next, but wtf, I'll detail it:

The freshly painted duct and flange, ready for installation:


The duct going onto the HSF:


Perfect fit!


The flange in place, to be fastened to the side cover:


Ready for installation. It's kinda obvious that the duct will pull cool air from the outside, directly to the HSF:


I know, some of you will say this isn't new... and you'd be right, I've seen machines from the mid-90's that had ducts like that from the factory. But that's not the point; the point is that we spend so much time with uber-coolers and watercooling, yet we forget one of the basic tenets of cooling, which is to make damn sure that the air going in and thru is as cool as humanly possible.

I think this qualifies.

Side panel with the flange installed:

Flange in place, fits nicely, and the paint on the screws even looks nice:


Side panel installed, duct on the flange:


View of the duct thru the back panel vent:


One complete HSF duct, ready for action:



The only "problem" with this is the other side panel fan's header is on the mobo next to the heat sink for the power control chips, near the back of the case, and it's a bitch to connect even without the duct. I should make an extender wire to make the job easier, but I shouldn't have to get inside there anytime soon.




And there you have a rig properly cooled. If any fans fail, there will be enough flow to keep operating, albeit at reduced efficiency.

The resulting airflow diagram:



Please forgive the shitty iPhone pics and shitty Paint.net diagrams, but I think you see the point, that the airflow is superior in this system and will ensure its reliability for a long time.

If I ever do get to upgrade, I'll probably do something like this:



Instead of the fans on the cooler tower, leaving a hotspot in the middle where the fan hubs are, by remotely locating the fans to the case body at either end of ducts, I get uniform airflow over the cooling fins, and ultimately superior cooling--and, if a fan fails, or both, it'll still work thanks to convection.

Thus concludes today's lesson on thermodynamics, your computer, and you.

Any questions?

let me bet the first to say....wft

Your posting advice about cooling while using a stock heatsink? All the improvements you made will not reduce your CPU's temperature as well as a good after-market heatsink. Believe me, I made this same discovery about 6 years ago with a Pentium 2.8c with a ducting system. This may be a fun project and good experience for modding, but its not worthy of a tutorial on cooling.

Direfox said:

Your posting advice about cooling while using a stock heatsink? All the improvements you made will not reduce your CPU's temperature as well as a good after-market heatsink. Believe me, I made this same discovery about 6 years ago with a Pentium 2.8c with a ducting system. This may be a fun project and good experience for modding, but its not worthy of a tutorial on cooling.

Click to expand...

10 degF drop is what I got, which isn't insignificant. You're right about the aftermarket HSF, but 'til I can afford one, which isn't right now, and I'm ready to overclock, this works for now.

Besides, the principles are sound, and worth noting.

So how about many conflicting airflows?

To better cool the video card I think that small section of holes next to the pci slots isn't going to help as much as popping off one or two of the card slot covers would.
this should help in your situation with your current fan setup...

*edit almost forgot about the fan on the side panel cooling off that section. The above advice in my post may not even be needed.

Also note that it's better to have all the airflow going in one direction inside the case. front to back or bottom to top are the best ways to have it.
As you have it now the rear fan location is going to pull in warm air from your power supply.

Ok I'll take this opportunity to ask this question, since I think it can benefit people with similar case setup.

Here's the diagram I'm about to work with. It's a Lian Li PC-P50 case, it has spots to place two (2) 120mm fans in front of the case, two (2) 140mm fans on top, and one (1) 120mm fan on the rear. I also have a 3rd party tower heatsink with push/pull configuration. What is the best way to orient all the case fans as well as the CPU heatsink for the best cooling? I also have a batmobile (or leaf blower) ATI reference cooler and a Corsair PSU (PSU config is not my concern). For side note I've already bought/obtained all the case fans needed for this case.

SOSTrooper, this might help you out:

This is my setup for cooling:


I know the cables look a mess! Gotta love the first generation antec 180B with no behind the mobo wiring at all

For scale this is another angle of the heatsink (Scythe Ninja):

or you could just get the FT02 or RV02 and have fans blowing everything upwards

I disagree with your application of the first law of thermodynamics Whilst the first law might be interesting, the only part of it terribly applicable to case cooling is that in differential form it'll tell you the higher your mass flow, the higher your heat loss. Also the more heat you create the more you dissipate, however the temperatures will also be higher... basically things that any case modder should know intuitively anyway

The more interesting thing is the fluid-thermodynamics involved, as most the cooling comes from either forced or unforced convection. The basics of that tell you bigger dissipation area (ie, bigger heatsink) the larger the rate of cooling, up to a point at least. Also that higher mass flow dissipates more heat, higher velocity dissipates more heat, turbulent flow dissipates more heat and a larger temperature difference dissipates more heat (ie. lower ambient air). So you want more fans, you dont want the fans fighting against each other, which will result in lower velocity (so dont have a front intake pointing straight at a rear intake) and also you want the most violently turbulent flow you can get (again, up until a point ).

I hate to break this to you OP - but your set up isn't that good. Too many fans fighting eachother - and no clear direction to the airflow. A big duct ruining any chipset cooling. This is why case designers moved away from fans on the door.. The guy with the 183 has the better idea but his cabling is a mess.. A setup more like that with clean cabling is the way to go..

GuyClinch said:

I hate to break this to you OP - but your set up isn't that good. Too many fans fighting eachother - and no clear direction to the airflow. A big duct ruining any chipset cooling. This is why case designers moved away from fans on the door.. The guy with the 183 has the better idea but his cabling is a mess.. A setup more like that with clean cabling is the way to go..

Click to expand...

Did you read my post?
Its one of the first gen 180's without any of the cable management behind the motherboard that was offered in later versions of the 180's and 183's. Sadly the cabling you see in that pic is the best I've been able to get it
I kinda gave up trying to make it look pretty when I realized it never will look good without the behind the mobo cable management.

Ducting the stock heatsink is not something that should be done. It's not just designed to cool the CPU.

Other then the OP's computer case and the date on the post, why does this feel like a necro'd post from like 5 years ago?


Adding a fan venting at the back or a second one at top would do a whole lot more for your coooling then that duct.

Tudz said:

Whilst the first law might be interesting, the only part of it terribly applicable to case cooling is that in differential form it'll tell you the higher your mass flow, the higher your heat loss.

Click to expand...

Wait, what? There are a couple of problems with this last statement. Firstly, I don't think that either you or the OP are really talking about the first law of thermodynamics, which (in layman's terms) is that energy is not created nor destroyed. Even if we start talking about differential changes in internal energy and entropy and work, it still doesn't seem to apply to computer cooling. Which equation are you referring to? As far as I know, the only truly relevant equation here is the basic convection heat transfer equation q/A = h(T1-T2).

Secondly, I don't think higher mass flows imply increased heat loss in this kind of situation. Isn't the heat load of the system fixed by the output of the processor? Even if we assume q to be variable, changing h only implies a change in q if the other variables are held constant.

Tudz said:

Also the more heat you create the more you dissipate, however the temperatures will also be higher... basically things that any case modder should know intuitively anyway

The more interesting thing is the fluid-thermodynamics involved, as most the cooling comes from either forced or unforced convection. The basics of that tell you bigger dissipation area (ie, bigger heatsink) the larger the rate of cooling, up to a point at least. Also that higher mass flow dissipates more heat, higher velocity dissipates more heat, turbulent flow dissipates more heat and a larger temperature difference dissipates more heat (ie. lower ambient air). So you want more fans, you dont want the fans fighting against each other, which will result in lower velocity (so dont have a front intake pointing straight at a rear intake) and also you want the most violently turbulent flow you can get (again, up until a point ).

Click to expand...

Again with implying that changing h affects heat load. Our heat load is fixed by the processor, so stop focusing on it. Increasing mass flow/velocity/turbulence (functionally this is the same thing. Just say Reynold's Number and be done with it) can only help us change our temperature differential... which is the whole point of cooling anyway.

Again: q/A = h(T1-T2)
q = heat load (watts)
A = area (m^2)
h = thermal coefficient, which is dependent on surface geometry and airflow
T1 = temperature of our source (the CPU)
T2 = temperature of our drain (our cooling air)

q is fixed, and unless we buy a new heatsink, A is also fixed. Let's also say that our cooling air temperature is fixed. That leaves us as having an inverse relationship between our CPU temperature and our thermal coefficient. If we want to lower our CPU temperature (yes!) then we increase our airflow. Again, heat load doesn't change.

If we use a duct to lower T2 - as the OP did - then we can get a lower T1. Other posters here have already discussed why this isn't the optimal solution.

Of course I'm making some serious assumptions, but I think they're valid. The first caveat is that processor efficiency is somewhat dependent on temperature. It's true that if you do a really good job of cooling your processor you can decrease your heat load substantially. It's usually less than 1W per 1°C, so I think it's safe to disregard this. Furthermore, workloads on the CPU are rarely constant (unless you're running Prime95 or something) but fortunately transients within a CPU resolve rather quickly back to steady-state. Also, assuming that cooling air temperature is fixed is a bit flimsy because we all know that the ambient temperatures within our case can fluctuate for any number of reasons. That's the reason why better case airflow and ducting is so desirable, but it really has nothing to do with thermodynamics.

The OP's post seems to be a lot more about airflow than thermodynamics and I'm fine with that despite the mistake in terminology. I hate to derail a thread but...

Did you read my post?
Its one of the first gen 180's without any of the cable management behind the motherboard that was offered in later versions of the 180's and 183's. Sadly the cabling you see in that pic is the best I've been able to get it
I kinda gave up trying to make it look pretty when I realized it never will look good without the behind the mobo cable management.

Click to expand...

Sorry. I didn't mean for it to sound like an attack. I am not so good at cabling either and some of those Antec cases are hard to cable. Just saying what I thought an ideal set up would look like..

pwrusr said:

Did you read my post?
Its one of the first gen 180's without any of the cable management behind the motherboard that was offered in later versions of the 180's and 183's. Sadly the cabling you see in that pic is the best I've been able to get it
I kinda gave up trying to make it look pretty when I realized it never will look good without the behind the mobo cable management.

Click to expand...

LOL I know exactly what you mean about the P180, I just switched from that to the FT02 I posted above... I am in heaven now.

SOSTrooper said:

Ok I'll take this opportunity to ask this question, since I think it can benefit people with similar case setup.

Here's the diagram I'm about to work with. It's a Lian Li PC-P50 case, it has spots to place two (2) 120mm fans in front of the case, two (2) 140mm fans on top, and one (1) 120mm fan on the rear. I also have a 3rd party tower heatsink with push/pull configuration. What is the best way to orient all the case fans as well as the CPU heatsink for the best cooling? I also have a batmobile (or leaf blower) ATI reference cooler and a Corsair PSU (PSU config is not my concern). For side note I've already bought/obtained all the case fans needed for this case.

Click to expand...

In cases such as this, the layout below has always shown to provide better temps over all, than any other setup I have tried.

BlueFireIce said:

In cases such as this, the layout below has always shown to provide better temps over all, than any other setup I have tried.

Click to expand...

Thank you, that's exactly what I need to know.

case cooling isn't rocket science. it's not about having as many fans as possible. it's about everything working together to flow the hot air out. generally speaking, a perfect setup is:

2 120mm fans intake at bottom front of case - standard placement.
1 120mm fan exhaust at top rear of case - standard placement.

the gpu and psu fans will *almost* make up for the 2nd 120mm intake. this is perfect because you want just a tad of positive pressure. plenty for all but the most extreme setups.

GuyClinch said:

Sorry. I didn't mean for it to sound like an attack. I am not so good at cabling either and some of those Antec cases are hard to cable. Just saying what I thought an ideal set up would look like..

Click to expand...

I didn't take it as a attack. I just got a bit mad that I couldn't do any better with cabling then what you see above. Antec really ticked me off when they came out with a much improved cable management system no more then 2 months after I got the case

zoglog said:

LOL I know exactly what you mean about the P180, I just switched from that to the FT02 I posted above... I am in heaven now.

Click to expand...

that's one case that's close to the top of my short list of cases that I may get.
the other is the corsair 600T. gotta wait a few months for that one to come out though

Kaldskryke said:

Wait, what? There are a couple of problems with this last statement. Firstly, I don't think that either you or the OP are really talking about the first law of thermodynamics, which (in layman's terms) is that energy is not created nor destroyed. Even if we start talking about differential changes in internal energy and entropy and work, it still doesn't seem to apply to computer cooling. Which equation are you referring to? As far as I know, the only truly relevant equation here is the basic convection heat transfer equation q/A = h(T1-T2).

Click to expand...

Ok my whole post was supposed to be somewhat joking, because the OP said...

Since the first law of thermodynamics states that temperature "flows" from hot to cold, this is critical to keeping our rigs cool.

Click to expand...

I guess my emoticons failed at bringing that across, but since you dont believe what I said...

The first law described mathematically is...

dU = dQ - dW

Where dU is change in internal energy, dQ is change in energy and dW is change in work. In layman terms this means what you said "energy is not created nor destroyed" because all those energies have to be in balance, but in mathematical terms it actually lets you calculate a lot. You can express it as a function of a spark ignition engine's crank angle, work in terms of pressure and volume (W = pdV) and internal energy as Cv*(T1-T2) and then rearrange it to calculate the power of an engine. So the first law isn't just some passing statement that energy is not created nor destroyed, its a tool used to calculate a lot of things (from temperatures to energies, mass flows and velocities). Newtons law of cooling is...

dQ = h*A*(T1-T2)

This can be substituted into the first law of thermodynamics if you so desire (noting that the first law is actually sums of energy, so it can have multiple Q, W and U terms).

You could also have Q as the heat released from components. W is mechanical work, of which not a lot goes on inside a computer except for maybe fans and pumps. U is internal energy which describes the energy contained in the air going into and out of the computer (it goes in cooler than it goes out as well as its velocity changes, so it gains energy as it goes through the computer).

Kaldskryke said:

Secondly, I don't think higher mass flows imply increased heat loss in this kind of situation. Isn't the heat load of the system fixed by the output of the processor? Even if we assume q to be variable, changing h only implies a change in q if the other variables are held constant.

Click to expand...

You stated the simplest form of newton's law of cooling, dQ = h*A*(T1-T2), it only really applies in that form to natural convection (ie, not fan forced). You can make it apply to other cases, as h is the heat loss coefficient. h is dependant on velocity of the fluid (ie, mass flow is important, you can't just have velocity with no net mass flow otherwise its just recirculating air and your delta T, (T1-T2), eventually becomes small and you lose your ability to reject heat.) It also depends on surface finish, turbulence, etc etc.

You'll also note that all those equations are in differential form, ie, they describe a RATE of heat loss rather than the actual heat loss. So they are talking in terms of heat loss per unit something. As such pretty much all of them can be expressed in terms of a mass flow rate (kg/s) and when you do, you realise any of those equations is dependant on mass flow rate.

Thermodynamics when you start talking about fluid flows is really really complicated and I'm well aware I dont understand it properly, anyone who tells you they do understand it fully is lying to you as you can go to any University with a fluids/thermo department and there are people still researching the fundamentals of it.

My original post was just meant to be silly because I've spent the past month or so reading thermodynamics and aerodynamics applied to gas turbine and spark ignition engines So often the statement comes up "from a first law balance we see that...(lots of complicated equations follow)". I didn't think anyone would actually care.

Yes, I know, I goofed, I got the wrong law, so sue me. Besides, the point is to shoe a decent setup and to initiate discussion, because there are so many setups that screw themselves with wrong airflow: watercoolers that push hot case air thru the radiator, or worse take the radiator exhaust and push it thru the case; not to mention aircooled rigs that screw up as well--mine is one of them because the exhaust CPU hsf exhaust is still pumped into the case and negatively affects overall cooling.

I understand well that higher ambient pressure improves heat transfer, but in my experience most positive-pressure setups tend to leave areas within the case with stagnant air that just keeps heating and heating, and tends to drive case temps up. My first blowhole was on a no-name case, and it didn't really work until I removed some of the intake fans and opened up some strategically-placed vent holes for fresh air in a negative-pressure arrangement.

I don't agree that I've got a problem with fans fighting each other, since my CPU temps dropped 10degF, and the mobo temp by about the same amount, which means the duct isn't occluding anything (in fact, the Asus docs pretty much state that chipset cooling comes from the CPU hsf flow). I am still thinking a "dam" from the duct to the upper rear corner of the case, spaced 1-2" off the mobo components, would encourage a bit more flow around the underside of the duct to the middle of the mobo, while allowing enough across the upper part to keep that cooler as well.

The vidcard is going to be an interesting issue, to say the least, tho again I'm thinking a duct pulling air in and back out without any mixing with case air. Of course, now we're in the realm of watercooling because that much ductwork might be too big a pain in the ass.

SOSTrooper said:

Ok I'll take this opportunity to ask this question, since I think it can benefit people with similar case setup.

Here's the diagram I'm about to work with. It's a Lian Li PC-P50 case, it has spots to place two (2) 120mm fans in front of the case, two (2) 140mm fans on top, and one (1) 120mm fan on the rear. I also have a 3rd party tower heatsink with push/pull configuration. What is the best way to orient all the case fans as well as the CPU heatsink for the best cooling? I also have a batmobile (or leaf blower) ATI reference cooler and a Corsair PSU (PSU config is not my concern). For side note I've already bought/obtained all the case fans needed for this case.

Click to expand...

Why not consider a duct, similar to what I drew up? That way NO HSF hot exhaust air gets into the case at all; nor will the mobo/vidcard heat affect the CPU cooling.

The stock 5870 cooler is already going to be exhausting air out the rear of the case and thus not really affecting the CPU cooling. Ducting a TRUE would be much more work than it would be worth I can assure you, the cooler is already quite efficient and I can almost guarantee you'll see betters temps by arranging the cooler efficiently with your case cooling (see diagram is BlueFireIce's post for more on this).

Tex Arcana said:

Why not consider a duct, similar to what I drew up? That way NO HSF hot exhaust air gets into the case at all; nor will the mobo/vidcard heat affect the CPU cooling.

Click to expand...

It's pointless to contruct a duct for a tower heatsink. Just have the rear fan blow into the heatsink and have the top fans both exhaust. Everything else intake.

Spare-Flair said:

Tex Arcana said:

Why not consider a duct, similar to what I drew up? That way NO HSF hot exhaust air gets into the case at all; nor will the mobo/vidcard heat affect the CPU cooling.

Click to expand...

It's pointless to contruct a duct for a tower heatsink. Just have the rear fan blow into the heatsink and have the top fans both exhaust. Everything else intake.

Click to expand...

I agree, a duct is pointless in most systems, Tex. In fact, a duct can hurt airflow for other, smaller components such as voltage regulators, northbridge/southbridge, and a ton of chips without their own heatsinks. I suggest you rename the thread.

What others have mentioned and shown here for their cooling setup is much more effective than ducts and a stock cooler.

On that note, I took the the more open space approach with more fans than I need so that I can run them through a fan controller to get a quieter PC (click for full size)...

Do you ever notice that since we are enthusiasts, we aren't concerned with .... well "cost efficiency".
So when a major OEM, is designing their lay out , for cooling... their concerns are stability, reliability, etc. Because the last thing they want is to have to service the system, remember American labor is a lot more expensive than Chinese labor. But we overclock you say, do we overclock to the point of the highest stable overclock, power consumption, heat and noise be damned? In my case going from 3.2ghz to 3.8 ghz on my chip, causes power consumption to more than double (q6600).
So it's commendable that he is trying to pull this off with Stock cooling, and using cheap but effective methods to cool his cpu.

Ducting does work, especially in cases where the major heat sources are far from fresh air sources.

@cyphernemo : Great job with your tagan black pearl. Did you experimented with bay fan devices?
I found that, a front fan pair, exhausting to just before the memory slots, helped Tremendously.
Also, having top fans blowing straight into the gpu also dropped temps by a bit less, 6c. This was done in a pcv1000, when I took a short respite from watercooling. In your case you would have to move your gpu up, if your board allows it...

avddreamr said:

Do you ever notice that since we are enthusiasts, we aren't concerned with .... well "cost efficiency".

Click to expand...

True, it might seem like that. But at the same time, if you compare enthusiasts with your run-of-the-mill gamer, the enthusiast knows how to save money. They know how to shop around for components, which components to skimp out on a bit and where to really invest. I do that all the time with friends and co-workers who need a gaming PC on a budget.

For example, the good ol' AC Freezer CPU cooler is a a value buy for cooling performance which I'd recommend for anyone on a budget. For myself, I might go with a Thermaltake, Scythe, or whatever flavor of the month. But I still avoid overpriced models and overpriced brand names like Zalman.

However, until now, I have yet to see an enthusiast who uses stock coolers, let alone takes photos of them. Now the OP knew he'd get crap for posting a thread titled "Today's lesson: Thermodynamics..." with a stock cooler on a hardware/overclocker forum like [H], didn't he?

avddreamr said:

So when a major OEM, is designing their lay out , for cooling... their concerns are stability, reliability, etc. Because the last thing they want is to have to service the system, remember American labor is a lot more expensive than Chinese labor.

Click to expand...

I don't know of any OEM system designers who read [H] forums. As you said, mostly enthusiasts here.

avddreamr said:

But we overclock you say, do we overclock to the point of the highest stable overclock, power consumption, heat and noise be damned? In my case going from 3.2ghz to 3.8 ghz on my chip, causes power consumption to more than double (q6600).
So it's commendable that he is trying to pull this off with Stock cooling, and using cheap but effective methods to cool his cpu.

Click to expand...

Everyone has their own motives for overclocking. For me, I overclock $300 and less CPUs to get similar performance to $999 and less CPUs. I also overclock to get the most out of my hardware. And although I can overclock my current CPU, the Core i7 920, to 4.1 GHz with it still being stable, I prefer to run it at 3.6 GHz since this is were the heat generated for each 100MHz changes from a gentle curve into an exponential spike. After all, I want my gaming PC to have the best of all worlds which includes running it fairly quiet.

avddreamr said:

Ducting does work, especially in cases where the major heat sources are far from fresh air sources.

Click to expand...

Ducting can work just fine in many scenarios. However, you're advising people to do this with stock coolers. Most people here don't run stock coolers. And most motherboards these days have voltage regulators that require their own heatstinks. Those heatsinks are meant to get some airflow from the CPU cooler as well. A duct system in the system you outlined would reduce airflow to secondary components in that PC. It's not going to hurt the PC, but ambient temperatures would surely rise.

Since this is supposed to be a guide on Thermodynamics, I strongly feel that it's counterproductive.

Were you just needing to cool the CPU, there's nothing wrong with a duct, and a duct should improve temps. But adding it to an entire PC that needs cooling for multiple components, each with their own, less-powerful air flow due to other fans and convention currents is a little off.

avddreamr said:

@cyphernemo : Great job with your tagan black pearl. Did you experimented with bay fan devices?
I found that, a front fan pair, exhausting to just before the memory slots, helped Tremendously.
Also, having top fans blowing straight into the gpu also dropped temps by a bit less, 6c. This was done in a pcv1000, when I took a short respite from watercooling. In your case you would have to move your gpu up, if your board allows it...

Click to expand...

Thanks. I have a bay fan at the top (it's from Scythe, but I had to mod it a little to get it where I like it). I have a second fan visible in front of the card slots, but I am thinking of adding a second bay facing inward from the bay to replace the middle-mounted one. Not sure yet.

As for dual bay fans in the font, that's not possible with the optical drive, memory reader, and fan controller there. Only an internally mounted bay fan pointed inward would be possible, but that's overkill. I always have a 120mm in front and behind the CPU cooler.

As for top fans pointed down into the GPUs, it would add a second direction to my airflow. I prefer one way airflows to minimize the need for higher fan speeds. With an easy path for air, less air pressure and/or vacuum is needed. Airflow is more efficient when you provide it with the path of least resistance. Complicating airflow with multiple directions and multiple bends just forces us to run fans faster. My GTX 295 already adds enough complication with an internal exhaust that I have to direct back down, out of the case.

There are two other reasons why I abhor cool air coming from the top. First is the additional accumulation of dust when the system is powered down. Second, is that you're fighting physics. The OP titled this thread "Thermodynamics", yet he shunned the concept. You did the same by suggesting I add fans to move cool air downward, into my case? Heated air rises. That's why we see some cases with vertical-mounted motherboards. That's why heatpipes need to be sealed and operate according to thermal conductivity and phase transition instead of convection. And that's why so many enthusiasts do what they do.