I've been putting together a list of components for a new computer for a while now. I'm getting stuck on which heatsink(s) to buy. The problem is that most (all) the reviews I can find for heatsinks assume a CPU hot enough to barbeque ribs. Since I'm going for a Core 2 Duo system which is a much cooler CPU than the reviews assume, I don't know which heatsink is ENOUGH. If I can avoid it, I really don't want darn near a kilogram of metal hanging off my motherboard. I know things like the ThermalRight Ultra 120 and Scythe Ninja (Infinity) are great, but do I really need that much heatsink? What's a reasonable heatsink for a non-overclocked E6700 CPU on a P965 motherboard?

On a related note, what about a reasonable heatsink on the northbridge? I've seen many threads (and one in particular here on SPCR) about the northbridge on P965 systems running hot. Shouldn't more emphasis be put on northbridge heatsinks? Why do we have kilogram-level CPU heatsinks but only 100-gram level NB heatsinks if their thermal loads seem to be approaching each other?

And then, there's the systemic heatsink problems. Again, I keep seeing things like:

- the CPU heatsinks won't mount properly because of interference from P965 motherboard components (memory sticks, the northbridge itself, and those little round capacitors around the CPU).
- sideways blowing CPU heatsink fans not cooling motherboard componenents (i.e., northbridge chips) properly.
- graphics card heatsinks and CPU heatsinks trying to take up the same space.

Of course, these types of things are motherboard specific. But, if anyone can point me to someplace which talks about these kinds of things systematically, I'd really appreciate it.

The best heatsink for a Conroe would probably be any heatsink that applies a lot of pressure. Or you could mod the rentention for several aftermarket heatsinks so that they do add more pressure.

Pushpin style heatsinks probably add the most pressure Stock. Kind of like the Stock Intel Heatsink.

the more effecient a heatsink, the less airflow required to keep it cool, which directly corilates to the rpm of the fan on the heatsink, and also the noise of the fan, or lack thereof.

If you use one of those monster oversized super effecient heatsinks like the ninja or Ultra 120, the slower you can run the fan on it, or possible even run it fanless.

I've got a Pentium M 1.73(533) with a thermalright Ultra 90 on it. I would have put the ninja if it fit on my motherboard. Its way overkill for the processor, but it allows me to run it completely fanless.

One of the reasons you'll see people putting Ninjas or other large tower heatsinks on their rigs is because larger heatsinks require less airflow to stay cool. Lower airflow requirements means lower speed fans which in turn means less noise. Hence the obsession with those heatsinks on this site. The real question you should be asking yourself is how much noise are you willing to put up with?



if you aren't overclocking the FSB, stock is adequate. Just remove the stupid foil covers if you go with an Asus board (on the southbridge too).



Mainly just the Scythe Ninja Plus rev B. I haven't heard of any others having mounting issues, though some force you to mount the fan to blow air upwards.

Only memory concerns I've read about have been just that - concerns. I don't recall seeing any that actually bumped into the memory (then again I don't read every post on every board out there)

Northbridge heatsink conflicts only occur when an aftermarket NB heatsink is used. And that's only when the CPU heatsink needs to be mounted such that the fan blows upwards.



easily rectified with a good airflow plan. Problems occur when people don't plan for it. Just remember a simple rule when planning - if the manufacturer put a heatsink on it, then it needs airflow.



Aftermarket GPU coolers only (and maybe some of the unique coolers on a few oddball models). Stock coolers are on the opposite side of the card as the CPU.

I think that's the whole point though - if something like a 92mm Nexus on an XP-90 were already capable of cooling, say, a (non-overclocked) Conroe while still running slow enough to be effectively inaudible in a given system, then there'd nothing to be gained by using an Ultra 120 or a Ninja, as it wouldn't be "more inaudible" whatever you did with it.

If the noise floor of rest of your system and your environment is so low that even the tiniest CPU fan noise troubles you, then a passive Ninja is probably the way to go, and if you want quietness *and* overclocking (or you're unfortunate enough to have a Prescott) your choice is similarly clear. Otherwise, the current mega-heatsinks seem like overkill and you might as well stick with something smaller, lighter and cheaper...

Exactly. For instance, if we take the massive heatsink phenomenom to the extreme and turn the whole case into a 14kg heatsink ala the Zalman TNN-300, SPCR gets CPU temperatures in the 40 C through 50 C range (fully passive). Specifically, using their CPUBurn run (for equivalence to the way they measure temperatures in heatsink reviews) gives a temperature of, essentially, 51 C regardless of added fans Zalman TNN-300 Fanless PC Enclosure System .

SPCR's review of the Scythe Infinity gives CPU temperatures in the range of 42 C through 53 C depending on fan voltage and orientation Scythe pushes towards Infinity .

Their review of the Thermalright Ultra-120 gives a range of 39 C through 52 C, depending on fan voltage Thermalright Gets Back on Top with the Ultra-120 .

The Zalman CNPS8000 ranges from 49 C through 60 C.

The Arctic Cooling Alpine runs from 53 C through 65 C.

The Spire Verticool II SP601B3 runs from 42 C through 69 C.

The Arctic Cooling Freezer 7 Pro runs from 39 C through throttle (about 75 C for the processor used).

And, thanks to a table in the Arctic Cooling Freezer 7 Pro review, the stock Intel HSF runs from 52 C through throttle, the Scythe Ninja w/ Nexus 120 fan runs from 40 C through 50 C, and the Thermalright XP-120 w/ Nexus 120 fan runs from 45 C through 68 C.

Except for the Zalman TNN-300, it looks like all of these measurements were done on the new socket 775 test system (P4-2.8 Prescott, 1 MB cache, 800 MHz FSB in 775 casing, Thermal Design Power is 84.0W, Maximum Power, as calculated by CPUHeat & CPUMSR, is 100.76W). The TNN-300 uses an Athlon 64-3000 (Venice core). So how those numbers compare is beyond me.

The point is that on the test system (Prescott), 75 C is bad (since the CPU throttles), and 52 C ought to be OK (since that's where the stock Intel HSF runs at full fan power). Since this is Silent PC Review, assume we want the lowest fan speed and a temperature no worse than OK. Then, the Ninja, Infinity, or Ultra-120 are all fine. But, where's that "OK" point with the Conroe Core 2 Duos like the E6700? What's the throttle temperature? If the TDP of the Prescott is 84W and the Core 2 Duos is 67W (about a 20% drop), can we assume all heatsink temperatures will drop about 20% (say, 10 C)? Is temperature change linear with power or is it something like a square or square root? Making the silly assumptions that the OK temperature and throttle temperature will be the same as with the older chips, and all heatsink temperatures will drop 10 C at any setting, then just about any heatsink will do, regardless of fan speed. The problem is, those ARE silly assumptions. But, I've seen nothing to give me any better data.

Sorry for the long rant.

After doing some more research, I found a bit more information. The spec sheet for the Intel® Pentium® 4 Processor 520 used by SPCR gives a TDP of 84W and a "Thermal Specification" of 67.7 C. The spec sheet for the Intel® Core™2 Duo Desktop Processor E6700 gives a TDP of 65W and a "Thermal Specification" of 60.1 C.

They define "Thermal Specification" as: "the maximum case temperature at the maximum Thermal Design Power (TDP) value for that processor. It is measured at the geometric center on the topside of the processor integrated heat spreader. For processors without integrated heat spreaders such as mobile processors, the thermal specification is referred to as the junction temperature (Tj). The maximum junction temperature is defined by an activation of the processor Intel® Thermal Monitor. The Intel Thermal Monitor’s automatic mode is used to indicate that the maximum TJ has been reached."

If I read that right, "Thermal Specification" is the maximum temperature that a given heatsink should allow the CPU to reach (as measured at that specified place which probably doesn't correspond to any of the temperatures available via the various tools out there). Since the Thermal Specification of the E6700 is 7.6 C cooler than the Prescott SPCR uses, does that mean the "OK" point I referred to in my previous post should be 7.6 C lower than it was with SPCR's test system? IOW, instead of the CPU staying near 52 C, with the Core 2 Duos we'd want the temps staying near 44 C? Or, am I totally mis-reading things?

to me, it just looks like your overthinking it way too much.

idunno about everyone else, but i just want it to work, and to work quietly. so i just buy the biggest, most effecient heatsinks i can fit on my motherboard and then i never have to worry about it.

i mean whats the harm in it? they all cost about the same price. some of the less effecient heatsinks cost more than the more effecient ones. So as long as it fits into your budget, and does what you want, who cares what the medium effeciency requirement for a specific CPU is?

Except I really don't want 960 grams (Scythe Infinity), 745 grams (Thermalright Ultra-120), or 665 grams (Scythe Ninja) cantilevered off the side of my motherboard (rated at 450 g) unless I really need to. Plus, if the CPU heatsink is too big, then it precludes some heatsinks on graphics cards and the northbridge.

I'd also like some kind of focus on heatsinks/temperatures on Conroe C2D processors because of some fairly fundamental problems with what constitutes the "temperature" of that specific CPU. There are a lot of threads out there with people wailing and gnashing their teeth because they're seeing CPU temperatures all over the map (even on the same machine at the same time, depending on the software they're using for the measurement). They have no idea whether their heatsink is messed up (perhaps interference from those capacitors I mentioned earlier or the heatsink's clamping pressure isn't great enough), their measuring tool is wrong, they're looking at the wrong measurement, their CPU is bad, or even if the temperature means bad things at all.

There is indeed lots of confusion out there about C2D temperatures, which displays amazing tenacity in spite of literally hundreds of posted explanations.

Most popular freeware (especially SpeedFan and Everest Home) does not currently support the on-chip digital temperature sensor (DTS) of the C2D. Some does: Core Temp, RMClock and TAT. Also Everest Ultimate ($30). These programs call this "core temperature".

Many motherboards use an on-board (vs on-chip) sensor for the CPU, which is completely useless. The Asus P5W DH is probably the worst offender here. Additionally the software that comes with those motherboards reports this misleading board temperature as CPU temperature (again Asus Probe 2 is a culprit). SpeedFan also uses this sensor.

The C2D DTS value is a negative number relative to the throttling temperature, which is a closely guarded Secret. In an attempt to be user-friendly, the various programs that read the DTS add its value to an assumed throttling temperature and report the resulting "actual" temperature. TAT and Core Temp use 85C, RMClock uses 97C, Everest Ultimate seems to use 85C but never reports a temperature higher than 81C, etc.

Your analysis above makes some incorrect assumptions. What matters here is the temperature rise induced by the thermal resistance of the cooling system (heat spreader, TIM, base plate, heat pipes, fins, air flow and turbulance). This resistance is nearly perfectly linear with the CPU power dissipation of the CPU for any given cooler. It is specified as a single value, *C/W, where lower is better, and 0.2 is a very good value.

An E6700 at stock frequency and voltage dissipates about 65W (an E6600 at stock is about 55W). If you assume 30C ambient inside the case (where the HSF is getting its air from) and want to maintain 15C of headroom below throttling, you can tolerate a rise of 40C, and therefore can use any heat sink with a thermal resistance less than about 0.6, ie, pretty much any heat sink.

Using a super-duper heat sink (Ninja, Ultra-120, Infinity, HR-01) will let you run with extremely slow fan speed or even passive. Using a very good heat sink (XP-120, SI-120, SI-128) will need a faster fan, but can still cool these CPUs at low speeds. These also have the advantage of pulling (or pushing) air through the motherboard components, particularly the VRM and north bridge, emulating the stock Intel cooler.

A long-winded way to say you don't need a tower with C2D unless you are overclocking or running passive. Something like an XP-120 is good enough. Make sure you get the appropriate mounting kit (typically sold separately).

Assuming that any heatsink with a thermal resistance less than 0.60 will be adequate and that the thermal resistance as tested on SPCR won't change just because the heatsink was moved to a different system, I went back and looked at all the socket 775 heatsinks I could find here. Since this IS SPCR, I filtered out heatsinks that throttled at 5v and had 5v sound measurements greater than or equal to 20 (arbitrary). Sorted by that sound measurement, I come up with the following (sorry about the formatting -- I haven't figured out if it's possible to do tables properly here):

HSF C/W MP @ 5V Weight (g) 5V Sound (dBA @ 1m)
Noctua NH-U12/Coolink U8-120-1600 w/ Nexus 120mm 0.32 800 16
Scythe Mine w/ stock fan 0.44 560 16
Noctua NH-U9/Coolink U8-92-1900 w/ Nexus 92mm 0.46 650 16
Thermalright HR-01 w/ Nexus 120 mm fan 0.28 525 17
Scythe Ninja w/ Nexus 120mm fan 0.29 665 17
Thermalright Ultra-120 w/ Nexus 120mm fan 0.29 745 17
Scythe Infinity w/ stock or Nexus 120 fan: Wide 0.29 960 17
Scythe Mine w/ Nexus 120mm fan 0.37 560 17
Scythe Infinity w/ stock or Nexus 120 fan: Narrow 0.37 960 17
Arctic Cooling Alpine 64 w/ Stock fan 0.42 486 17
Thermalright XP-120 w/ Nexus 120 fan 0.47 370 17
Noctua NH-U9/Coolink U8-92-1900 w/ stock fan 0.35 650 18
CoolerMaster Hyper 48 w/Stock Fan 0.53 864 18
APack ZeroTherm BTF90 (Copper) w/ Stock fan 0.26 678 19
APack ZeroTherm BTF80 (Aluminum) w/ Stock fan 0.28 458 19

I haven't seen anything about the actual energy consumption of the E6700 at full bore (i.e., when CPUBurn is running). Assuming the highly unlikely value of 100w for that and plugging that back into the thermal resistance equation gives a *C/W cuttoff of 0.4 instead of 0.6. That change drops the following heatsinks off the list:

Arctic Cooling Alpine 64 w/ Stock fan
Scythe Mine w/ stock fan
Noctua NH-U9 / Coolink U8-92-1900 w/ Nexus 92mm
Thermalright XP-120 w/ Nexus 120 fan
CoolerMaster Hyper 48 w/Stock Fan

Assuming the difference between 16 and 19 dBA outside a case is negligible and that the next most important thing to sort on is thermal resistance, that gives the following list:

HSF *C/W MP @ 5v
APack ZeroTherm BTF90 (Copper) w/ Stock fan 0.26
Thermalright HR-01 w/ Nexus 120 mm fan 0.28
APack ZeroTherm BTF80 (Aluminum) w/ Stock fan 0.28
Scythe Ninja w/ Nexus 120mm fan 0.29
Thermalright Ultra-120 w/ Nexus 120mm fan 0.29
Scythe Infinity w/ stock or Nexus 120 fan: Wide 0.29
Noctua NH-U12 / Coolink U8-120-1600 w/ Nexus 120mm 0.32
Noctua NH-U9 / Coolink U8-92-1900 w/ stock fan 0.35
Scythe Mine w/ Nexus 120mm fan 0.37
Scythe Infinity w/ stock or Nexus 120 fan: Narrow 0.37

Sorting by weight gives:

HSF Weight (g)
APack ZeroTherm BTF80 (Aluminum) w/ Stock fan 458
Thermalright HR-01 w/ Nexus 120 mm fan 525
Scythe Mine w/ Nexus 120mm fan 560
Noctua NH-U9 / Coolink U8-92-1900 w/ stock fan 650
Scythe Ninja w/ Nexus 120mm fan 665
APack ZeroTherm BTF90 (Copper) w/ Stock fan 678
Thermalright Ultra-120 w/ Nexus 120mm fan 745
Noctua NH-U12 / Coolink U8-120-1600 w/ Nexus 120mm 800
Scythe Infinity w/ stock or Nexus 120 fan: Wide 960
Scythe Infinity w/ stock or Nexus 120 fan: Narrow 960

Now, I still have to determine qualitative factors that might change those rankings and figure out if those heatsinks will physically fit on P965 motherboards (and an Antec Solo case).

I would deduct weight as a criteria. None of those heatsinks, even the heaviest ones, have had any real reports of issues from their weight.

In place of weight, I would add a couple of things:

Price.

And "flexible installation". Having a tower that can point in more than one direction is a very handy thing. Or, using you motherboard's socket orientation plus the heatsink installation instructions, eliminate any towers on your list that point in the wrong direction.

IMPO it's only a matter of time; the Infinity hasn't been out very long, so there's not really any large pool of long-use examples to draw statistically significant conclusions from. The stock retention mechanism for the Infinity is just too flimsy for 1kg+ of weight. And indeed the whole point of this thread is that with cooler-running CPU's now such mammoth heatsinks aren't neccesary to achieve adequate cooling. For peace of mind if nothing else I would consider weight one of the chief criteria.

The depressing thing about my list, above, is that with my cutoff of *C/W = 0.4, every single non-tower (i.e., blowing towards the motherboard) drops out. Thus, no way to use the CPU heatsink to improve airflow across those hot MOSFETs and Northbridge. I ran across a few posts around the web that the Thermalright SI-120 gives better clearance from motherboard components and runs 3-4*C cooler (inside a case -- not on an open testbench) than their XP-120. So, I thought I could finagle it onto the list. Unfortunately, I can't find it anywhere. It seems to have been replaced by the SI-128. That, unfortunately, weights 510 grams, is taller, wider, and seems to have more tightly packed fins (probably worse performance at slow air speeds). Heck, I can only find one place that claims to still carry the XP-120.

BTW: I found ONE reference that specified the maximum power dissipation for the E6700. It said 95 watts instead of the 100 watts I assumed. I have no idea where they got that number. Using that instead of 100 watts raises the allowed *C/W from 0.4 to 0.42.

OK, so you have this one number, that you have no idea where they got it from, and you are basing your whole heatsink decision-making process around it? Also, are you planning to run CPUBurn 24/7, and if not, why is it the yardstick by which you are choosing your heatsink?

In normal maxed-out use, I would expect the E6700 to dissipate a little more than 65W, but not a lot more (certainly not almost 50% more).

The whole point of the TDP (Thermal Design Power) is that this is the wattage your cooling mechanism has to be able to cope with; my basic point is that you are using the wrong number for heat dissipation.

Also, you can undervolt the CPU to reduce heat dissipation.

Also, it's not entirely true that tower heatsinks can't cool MOSFETs, you can put a downward angled shroud on the back of the heatsink to direct air towards them; the northbridge would probably need its own cooling solution anyway.

Assume the mobo is designed to use both the old P4s and the new Conroes. The P4s use lots of power! Yet the motherboard is designed to work at those power levels reliably with no special cooling provisions for the MOSFETs. When using the much cooler-running Conroe, the MOSFETs will likewise run much cooler. And since they needed no special cooling at far higher power levels...

Don't sweat the MOSFET (poetry intended).

Sorry for jumping in but saves starting a whole new thread........

Can anyone suggest low profile coolers for an Allendale/Conroe in an Antec NSK2400 case???

Never used an Intel before but from the little I have read it appears that the stock cooler is rather noisy.

The height of my choice of case rather restricts hsf choices.
Any suggestions?

But, that's my whole point in this thread. No one seems to be doing any testing of heatsinks with Conroe in mind. In order for anyone to make any decision on a heatsink for a Conroe processor, they have to make a whole series of totally unjustified, pull-it-out-their-b*t, assumptions. Exactly like I'm doing here. I don't WANT to do this. I just have no alternative. There's no real information out there to use. Essentially, I'm just trying to extrapolate from fry-an-egg processors to Conroe.

As for CPUBurn, I'm using it as a comparison point (as well as a guess at *C/W (MP) instead of *C/W (TDP)) because that's what SPCR uses and all the data I'm referring to comes from them.



I'm no expert at this. But, from Intel's spec sheet for the E6700:



The last four words are what I'm worried about: "under normal operating conditions." To me, that means browsing the web and writing a document. But, I need a heatsink that will handle everything I'll do. On my current system, I've no problem "under normal operating conditions." But, when I do things that stress the system, I quickly run the processor temp up to 75 degrees (where it always stabilizes, so I assume it's throttling). I don't want to make that same mistake with a new system. SPCR uses http://www.cpuheat.wz.cz/html/Pentium4.txt to come up with a value to approximate the "W" in their *C/W (MP) formula to correspond with what they think CPUBurn is doing. That site doesn't seem to have anything on Conroe processors. If anyone can come up with a more realistic number for the Maximum Power Thermal Dissipation for a Conroe, point me to it. I'd like it to be much smaller than 95W, too. But, if I have to make unfounded guesses to make decisions (since no one seems to be doing any actual testing and measurement), then I'm going to use worst case numbers.

Probably true. But, I see people saying that since their Conroe runs so cool, they'll just stick a Ninja in there and run everything passive: passive CPU heatsink, passive Northbride, even a passive graphics card (heck, I was thinking about it). Except, motherboards are designed with the assumption that there will be SOME air flow. Specifically, since Intel's stock heatsink blows towards the motherboard, I'd guess they're assuming some air flow down over those MOSFETS. Here's a not-very-applicable example of what MOSFET temperatures can do without airflow. It's a very different situation, but it's something to consider. So, if it's at all possible, I'd like a CPU heatsink that pumps air towards the motherboard without my having to set up shrouds, ducts, and extra fans.

Jeez, make up your mind!

At the start, you said you were building a system around a non-overclocked Conroe, and we all said that a Ninja passive would be good, or an XP-120-class cooler with a fan would be good. What else do you think is necessary to know????

As for the MOSFETs, that depends on the motherboard. A P5W or DQ6 does not need any cooling of the MOSFETs for a Conroe, whereas a P5LD2 might, but probably doesn't.

A Conroe at stock frequency and voltage dissipates at most 65W, and is very unlikely to overheat the MOSFETs.

While difference decreases with lower rpm it's still little better than SI-120.
http://babelfish.altavista.com/babelfish/trurl_pagecontent?lp=de_en&trurl=http%3a%2f%2fwww.hartware.de%2freview_597_8.html
(PWM temperatures aren't so relevant, open testbench)

about Speedfan

latest version of Speedfan does report Intel Core temp

SpeedFan 4.32 beta is online. This beta adds full support for INTEL CORE internal temperature readings. I had to rewrite some code that handles CPU LOAD reporting. Let me know if everything works fine for you and if you have a CPU that should be supported, but that still isn't. I would like to receive reports from different OSs (like Win2K3, Windows 9x and Windows NT).

http://www.almico.com/sfbetaprogram.php

The first line in the main body of the SPCR home page says:



That's exactly what I want. And, since the Conroe TDP is so different from other recent Intel chips, I'd like that stuff to also include Conroe information, not P4 information. Things like "we all said," "might, but probably doesn't," and "very unlikely" aren't what this site is about. If they were, then the site wouldn't need to exist at all.

Regardless, it's all moot. I found a site that actually went out and counted the number of teeth on a real, live horse instead of depending on Aristotle or Archimedes to provide the philosopically correct answer:

AMD’s Response to Intel Conroe: Energy Efficient Athlon 64 X2 CPU

In a nutshell, an E6300 (TDP 65W) drew the following amouns of power:

Idle: 26W
Video Playback: 29W
3DMark: 40W
CPU Burn: 44W

So, with that in mind, I'll agree that 65W is a decent limit for the amount of power a Conroe chip will draw (heck, it appears to be an extraordinarily conservative (i.e., safe) limit). That puts us back to a *C/W of 0.6 and just about any heatsink on the planet will be thermally adequate. Soundwise, if the heatsink's stock fan can be replaced with a Nexus 120mm fan, then that'll do.

Now, if someone would actually measure those MOSFET temperatures with different CPU heatsinks in a closed case, that'd be great.

you called?
http://www.madshrimps.be/gotoartik.php?articID=389

OK, I see now what's happened here; you've slightly misunderstood the definition of TDP and the extreme heat problems of the Netburst era have made you very cautious with regard to the heat dissipation and how you will deal with that.



When the TDP guideline says "under normal operating conditions" what it means is that this is the maximum you could expect to see with the case in an average temperate/continental climate, average case ventilation and using the most power-hungry real-world application at stock speeds and voltages (ie not an artifical stress test like Prime95 or a power virus).



It's lower, but apart from that not much has changed; the TDP is the number for the highest-clocked chip in the range, and lower-clocked models will run correspondingly cooler, as you have found with those E6300 numbers.

who cares what temp the mosfets are, as long as its stable.

were not here to OC, were here to achieve a quiet stable computer. the mosfets will be fine being 1-5c hotter with a tower heatsink.

i dont really think weight should be an issue, never heard a single person say "man this ****** tower heatsink just wrecked my cpu/motherboard because it was so heavy when i went to move it".

plus with the ninja, the majority of the weight is in the base, so the cantelever effect is minimal. The only time i could possibly see it being an issue is if you were shipping your computer via UPS or somthin similar. In which case you should have packed it with some sort of foam or bubble wrap anyhow to keep anything inside the computer case from moving around, including the cpu heatsink.

Heres what i think should be on your list of considerations:

Price
Physical Size Compatability

thats it. If it fits in your budget, and fits on your motherboard with your system including all your expansion cards etc, then who cares about all the extra random stuff?

Remember, these systems usually have a rear exhaust fan, plus the fan that's in the PSU, which is often located just above the CPU. All the air in the system gets pulled into that corner of the case and exhausted. There is ample air movement around those hot parts, and the Ninja's non-directional, open fins cause little restriction. I think you are worrying too much. Also, the Ninja is nicely balanced, with most of its weight is in the base as Aris just said. It feels very secure and stable on the motherboard.

Ahem!

It's an interesting comparison, but the question is, how much is too much? I think both Aris and I were suggesting that the Ninja looks a lot more top-heavy than it actually is. If I had any doubts about it, I wouldn't use it.

Thanks for the link. I'd been to your site several times before while looking for information. Browsing through your list of reviews, it looks like most of them are on an AMD platform, some are Intel, some are open case, some are closed case. Assuming MOSFETs in an AMD system will behave similarly to those in an Intel system, it looks like their temperatures parallel the CPU's pretty closely no matter what the heatsink is. Again, I was just browsing through the results, but the worst case I saw was a 11*C difference between CPU and MOSFET temperature (most were much closer). So, that lays my MOSFET temperature fears to rest.