Nine Big Air Coolers For Intel's Haswell CPUs, Reviewed

[NucDsgr]

"a quad-core Haswell-based processor at 22 nm requires more cooling than a six-core Sandy Bridge-E CPU at 32 nm, even though its lower power "

It sounds like a anthers case of Intel using their lousy thermal compound on Haswell retail processors. Bet the samples released to the media does not feature this lousy paste.

Remove the cover plate, scrap off the compound from the bare silicon, mount the processor on the motherboard sans coverplate , apply Arctic Silver Paste, and see how well the thermals are.

Real enthusiasts are not afraid of taking cover plates off. Been doing it for years. The Sandy Bridges and Ivy Bridges also have lousy paste.

 

[RedJaron]

i don't know... this h100 was a pretty good deal at $50 new.

If I could always count on getting such a great deal, I'd do it. But seeing as how I can consistently get a Noctua for less means I'll stick with air for a while.

I don't think anyone is discounting this... i think the issue is they're asking for more money for both the haswell cpu and z87 mbs are more expensive as well, and because of the heat issues you get pretty much zero NET improvement over SB or IB. seriously, look at the benches we're talking about 1 or 2 fps difference in pretty much every game. What's the point of improving the IPC if it nets you nothing in practical experience?

Fair point. I'm not saying Haswell isn't a bit of a disappointment for desktop users, I'm just trying to point out it's not as bad as many try to make it out to be. As for price, Haswells and 8-series boards are selling close to the same amount IB did when it was released. As for gaming fps gains, those have been marginal for years ( Nehalem to SB, SB to IB, etc. ) Upgrading isn't all about IPC. It's about better memory controllers, increased instruction support, connectivity lanes, lower power and thermals, and a lot of other stuff as you go from one platform/generation to the next. Admittedly Haswell muffed up on some of that list. No, it doesn't make a lot of sense for someone to upgrade to Haswell from IB or SB, but if you're on Nehalem or older, there's not a lot of reason to get an IB over Haswell if you want a new box.

 

[computertech82]

I have no issue using the DEFAULT heatsink for my 4570 cpu. 40 celsius.

 

[mapesdhs]

Real enthusiasts are not afraid of taking cover plates off. Been doing it for years. The Sandy Bridges and Ivy Bridges also have lousy paste.

Problem is that voids the warranty. I had an i7 870 die after 2 years of use; got it replaced
under warranty no problem. Being able to do that with a 6-core which cost 3X more would
be even more important, not an ability I'd be willing to lose. I can only conclude that the
'real' enthusiasts are the ones with limitless budgets. :}

Ian.

 

[Calculatron]

@ mapesdhs - The TRUE you are referring to is the Thermalright TRUE Spirit, right? I believe there are some sites that use that, and similar models, as a basis for heatsinks for terms of costs and performance.

 

[JimmiG]

JimmiG writes:
> My 4770K with HT enabled @ 4.3 GHz/1.214V runs about 85C in Linpack. ...

Ouch, for 4.3 that sounds kinda hot. My 2700K @ 5.0 stays under 80C (1.424V in CPU-Z).

Pretty normal for Haswell as you can see in the review. They got 65C above ambient at 4.5 GHz / 1.25V, and I'm at <60C above ambient at 1.214V. I'm not sure about my case temp since I can't measure that, but room temp is around 25C. Case might be ~5C above that maybe.

Anyone who claims lower temps than that on air either got an extremely rare CPU or they aren't stress-testing it properly. Linpack and AIDA64 FPU-only are the best and you have to use all 8 virtual cores with HT enabled for maximum heat.

I'm kind of tempted to get the Corsair H110 since I have room for it in my case, but given the small difference in temps between the air coolers tested, I'm not sure it would make *that* much of a difference. With the heat building up and getting stuck under the IHS, a high-end cooler never really gets utilized fully with Haswell.

 

[mapesdhs]

Calculatron writes:
> @ mapesdhs - The TRUE you are referring to is the Thermalright TRUE Spirit, right? ...

Not quite, I meant the standard Thermalright U120 Extreme Rev C (the previous revision
is basically the same, just a different mounting system). The TRUE Spirit was more expensive,
not sure why.


> ... I believe there are some sites that use that, and similar models, as a basis for heatsinks
> for terms of costs and performance.

Yes, the standard TRUE became a widely used baseline for HS comparisons, a bit like the HS
equivalent of, can it run Crysis? 😀

Ian.

 

[mapesdhs]

JimmiG writes:
> Pretty normal for Haswell as you can see in the review. ...

Sadly yes. :\


> ... a high-end cooler never really gets utilized fully with Haswell.

I've seen people elsewhere refer to 'chiller' coolers, giving load temps below 10C. Do you
know anything about those?

I suppose the only real step up beyond an H100 or high-end air cooler is a custom water loop.
Think I need to start learning about home-done water cooling this year, especially if we're going
to keep getting summers like this one (room air temp is 28C atm; glad I don't live in London,
they must be all spit-roasted by now). Today I had a strawbery Cornetto for breakfast. 😀

Ian.

 

[Crashman]

JimmiG writes:
> Pretty normal for Haswell as you can see in the review. ...

Sadly yes. :\


> ... a high-end cooler never really gets utilized fully with Haswell.

I've seen people elsewhere refer to 'chiller' coolers, giving load temps below 10C. Do you
know anything about those?

I suppose the only real step up beyond an H100 or high-end air cooler is a custom water loop.
Think I need to start learning about home-done water cooling this year, especially if we're going
to keep getting summers like this one

The difference between internal die temperature and cooler base temperature is always high on Haswell. Many of the guys who report low full-load CPU temperature are likely getting external readings or faulty readings. The spread between die and cooler temperature gets worse as you lower base temperature, hence the charts that show coolers with twice as much capacity getting maybe 15% better temperatures.

What am I getting at? The biggest coolers are already running only a little above ambient (maybe 10°), yet the CPU core is still 65° over ambient. If you use big liquid to get the "cold plate" to 5° over-ambient, you're probably only going to see a 2-3° drop in CPU temperature. The next step is sub-ambient cooling.

 

[Novuake]

JimmiG writes:
> Pretty normal for Haswell as you can see in the review. ...

Sadly yes. :\


> ... a high-end cooler never really gets utilized fully with Haswell.

I've seen people elsewhere refer to 'chiller' coolers, giving load temps below 10C. Do you
know anything about those?

I suppose the only real step up beyond an H100 or high-end air cooler is a custom water loop.
Think I need to start learning about home-done water cooling this year, especially if we're going
to keep getting summers like this one

The difference between internal die temperature and cooler base temperature is always high on Haswell. Many of the guys who report low full-load CPU temperature are likely getting external readings or faulty readings. The spread between die and cooler temperature gets worse as you lower base temperature, hence the charts that show coolers with twice as much capacity getting maybe 15% better temperatures.

What am I getting at? The biggest coolers are already running only a little above ambient (maybe 10°), yet the CPU core is still 65° over ambient. If you use big liquid to get the "cold plate" to 5° over-ambient, you're probably only going to see a 2-3° drop in CPU temperature. The next step is sub-ambient cooling.

Both points are perfectly valid... Competittion makes things happen faster. YOu are debating the on same side of the table... 😀

 

[Crashman]

I suppose the only real step up beyond an H100 or high-end air cooler is a custom water loop.
Think I need to start learning about home-done water cooling this year, especially if we're going
to keep getting summers like this one

The biggest coolers are already running only a little above ambient (maybe 10°), yet the CPU core is still 65° over ambient. If you use big liquid to get the "cold plate" to 5° over-ambient, you're probably only going to see a 2-3° drop in CPU temperature. The next step is sub-ambient cooling.

Both points are perfectly valid... Competittion makes things happen faster. YOu are debating the on same side of the table... 😀

Partly, I'm just saying that even a custom water loop isn't going to help a lot with Haswell. One really needs to go below-ambient or fix the heat spreader issue.

 

[Novuake]

Partly, I'm just saying that even a custom water loop isn't going to help a lot with Haswell. One really needs to go below-ambient or fix the heat spreader issue.

Which is easier than most people think... But then again, ALWAYS a risk involved.

EDIT : The AGAIN. Intel should have done it properly in the first place.

 

[Crashman]

Partly, I'm just saying that even a custom water loop isn't going to help a lot with Haswell. One really needs to go below-ambient or fix the heat spreader issue.

Which is easier than most people think... But then again, ALWAYS a risk involved.

EDIT : The AGAIN. Intel should have done it properly in the first place.

I'm not so sure about the easy end-user solution (heat spreader removal) being a good one though. A heat spreader has more surface area than the die itself, and solder is an awesome conductor. Because the die is so small and solder is so much better than paste, the larger surface area of a soldered-on heat spreader might be better than direct sink installation upon the die.

I know of a way to find out. Get some processors and some low-heat solder, measure the processor's full-load temps under a fixed set of conditions, measure it again with the heat spreader removed, then clean out the heat spreader and add solder before retest. I think we can just preheat the CPU so the die doesn't crack when dipped into the solder-filled heat spreader 😛

 

[flong777]

Oftentimes in engineering, the simplest solution is the best solution. In this case, the direct approach of highly efficient air coolers like the NH-D14 appears to be the overall best cooling solution in general. While custom water loops are more efficient, they really belong to the boutique, high--end builder that is focused more on producing an artwork than on actual performance increases.

In the reviews that I have read, even the best custom water loops only beat the NH-D14 and similar coolers by 3-5 degrees Centigrade - that really isn't a lot if you consider the overall operating temperatures of CPUs in general.

What is even more troubling about the expense and complexity of water loops is that they do not direct heat away from other components - only the CPU unless you add connections for the GPU. High-end air coolers DO direct heat away and out of the case from all surrounding components which is another benefit that is rarely considered in testing (some tests do look at this).

I like the prefab water loops like the Corsair H100i but overall it doesn't cool other components in the case (though you can set it up to vent hot air out of the case). It does a great job cooling the CPU but not other components.

So we can see that the simplest, most direct solution of air cooling is the best bang for your buck. As Crashman noted, water loops are limited by the same thing air coolers are, they both cannot go below the ambient temperature. This is what hurts water coolers (though they are much more efficient at directing heat away from the CPU because the water transfers the heat more efficiently than air). Still 5 degrees Centigrade just doesn't seem to be worth the trouble and expense to switch to water cooling.

 

[Novuake]

I'm not so sure about the easy end-user solution (heat spreader removal) being a good one though. A heat spreader has more surface area than the die itself, and solder is an awesome conductor. Because the die is so small and solder is so much better than paste, the larger surface area of a soldered-on heat spreader might be better than direct sink installation upon the die.

I know of a way to find out. Get some processors and some low-heat solder, measure the processor's full-load temps under a fixed set of conditions, measure it again with the heat spreader removed, then clean out the heat spreader and add solder before retest. I think we can just preheat the CPU so the die doesn't crack when dipped into the solder-filled heat spreader 😛

Simple enough, as I said... 😀 I do not like the whole shiv/spacer debate much either, I can not imagine the shiv can conduct the heat as well as even paste does since you now have double the mating surfaces...

+ Biggest mistake you can make is glueing the IHS back with no room to breath... LOL. Saw some photos of a CPU ripped apart( exaggerating maybe a tiny bit ) by the contractions.

 

[Crashman]

Simple enough, as I said... 😀 I do not like the whole shiv/spacer debate much either, I can not imagine the shiv can conduct the heat as well as even paste does since you now have double the mating surfaces...

+ Biggest mistake you can make is glueing the IHS back with no room to breath... LOL. Saw some photos of a CPU ripped apart( exaggerating maybe a tiny bit ) by the contractions.

You'd only want enough solder "fill" to cover the top. I still think that solder can perform so much better than paste that there's a chance the extra layer could be more than negated by the wider contact area.

 

[Novuake]

Simple enough, as I said... 😀 I do not like the whole shiv/spacer debate much either, I can not imagine the shiv can conduct the heat as well as even paste does since you now have double the mating surfaces...

+ Biggest mistake you can make is glueing the IHS back with no room to breath... LOL. Saw some photos of a CPU ripped apart( exaggerating maybe a tiny bit ) by the contractions.

You'd only want enough solder "fill" to cover the top. I still think that solder can perform so much better than paste that there's a chance the extra layer could be more than negated by the wider contact area.

I understand. 😀 Was just saying that its something to consider. But yeah, numbers should be much better but truly how many Ivys are even capable of exceeding 4.8GHz stably, even with great temps. Temps just are not the only limiting factor.
Trigate transister, 22nm, all of these are new to Ivy and complicate things compared to Sandy, so overclocking it to the same heights are expecting a bit much.

 

[masmotors]

im cooling my 4670k with a 620 h2o some of those fans look real cool which would beat my 620

 

[dgall]

Legitimate question though- why isn't the first section of the article titled "Picking the Best Manufacturer-Sent Limited Selection Sample Cooler for Haswell?"

I believe Tom's has a certain responsibility to point out that the contents of their articles are dictated by manufacturer samples, more than showing us a legitimate selection of top products.

 

[Novuake]

Legitimate question though- why isn't the first section of the article titled "Picking the Best Manufacturer-Sent Limited Selection Sample Cooler for Haswell?"

I believe Tom's has a certain responsibility to point out that the contents of their articles are dictated by manufacturer samples, more than showing us a legitimate selection of top products.

READ the thread.

 

[Djentleman]

Didn't mention the crappy thermal paste they put under the lid of the cpu's.
Delidded mine and got about a 10c decrease.

 

[mapesdhs]

... One really needs to go below-ambient or fix the heat spreader issue.

Where to begin the learning curve? Is there a good site resource you know of for getting started with
that type of cooler? I read in a post the other day someone referring to a 'chiler', with which they had
their CPU at 0C.

Custom water loops is daunting enough, but for meddling with below-ambient I'd really want to
properly read up on how it's done, etc., before trying it out.

Ian.

 

[Crashman]

Legitimate question though- why isn't the first section of the article titled "Picking the Best Manufacturer-Sent Limited Selection Sample Cooler for Haswell?"

I believe Tom's has a certain responsibility to point out that the contents of their articles are dictated by manufacturer samples, more than showing us a legitimate selection of top products.

Tom's Hardware set a fairly high bar for entry. Some coolers failed to reach that bar. One would think that, given such a high bar, companies would want to put forth their best efforts. And several did.

 

[raykat]

It seems to me that from day one Intel designed Haswell more for laptops than for serious system builders i kinda feel like we got left overs mmmm

 

[Y0GI]

I don't understand the results, because the Phanteks PH-TC14PE had the lowest Delta T (64* above ambient) and yet the Noctura was selected as the best of the coolers tested? What am I missing?

Yogi