SPCR’s 2010 CPU Heatsink Test Platform [Updates: 10 April & 31 May]

Our CPU heatsink test platform has been updated with a hotter, more power hungry CPU. Eight of the top heatsinks we tested previously get put through the wringer once again. Some continue to excel while others drop to the wayside.

January 25, 2010 by Lawrence Lee & Mike Chin

POSTSCRIPT 2: TEST PLATFORM FOR SMALLER HEATSINKS 31 May 2010 – see page 6

 

POSTSCRIPT: O THE TESTING LIFE! 10 April 2010 – Motherboard replaced – see page 5

For many of us, the journey into silent computing began with the headaches caused
by the rattling and droning of the dreaded stock CPU cooler. Whether you had
an Intel or AMD processor, the reference heatsink/fan was guaranteed to be loud
and inefficient and suitable replacements were few and hard to come by. It wasn’t
until Zalman launched their famous 7000
series that an average end-user could actually get better cooling with
less generated noise at a reasonable price.

Today the situation is much improved with a wide array of coolers to choose
from. Quiet and loud, big and small, extravagant and simple, there is a smorgasbord
of heatsinks out there which makes reviewing them as important as ever. With
so many choices it’s paramount that we be able to judge relative performance
as objectively as possible while taking into consideration real life application.

CPU Heatsink Test Platform, 2006-2009

• Intel
  Pentium D 950 Presler core, C1 stepping. TDP of 95W.
• Asus P5Q-EM motherboard.
  A microATX board with integrated graphics and short solid-state capacitors
  around the CPU socket, and a diminutive northbridge heatsink for maximum compatibility.
• Intel
  X25-M 80GB 2.5″ solid-state drive.
• 1GB
  of Corsair XMS2 DDR2 memory. 2 x 512MB PC2-8500.
• FSP Zen 300W
  fanless power supply.
• Arctic Silver
  Lumière: Special fast-curing thermal interface material, designed
  specifically for test labs.
• Nexus 120 fan (part of our standard testing methodology; used when
  possible with heatsinks that fit 120x25mm fans)
• Custom-built, four-channel variable DC power supply, used to regulate
  the fan speed during the test.

The previous CPU heatsink test board, the Asus P5Q-EM.

Our current CPU heatsink test platform consists of a Pentium D processor on
a mATX LGA775 motherboard. A quick-curing thermal compound is used, and if possible,
a reference 80/92/120mm Nexus fan. The fan is attached to a custom fan controller
to vary the fan speed so we can obtain a good cross-section of performance with
regard to airflow and noise. The platform is placed horizontally on a wooden
structure, unenclosed. Except for a change in the motherboard last year, our platform
has not really changed in several years. With 2010 upon us and the Pentium D
950 passing its 4th birthday, it was decided that the platform was due for an
update.

Component Changes?

The Intel Pentium D950 is not as power hungry or hot as today’s high-end processors and its die size is smaller. As a result many of the heatsinks we’ve tested appear to be very close to one another in performance, sometimes making it difficult to conclusively say whether one is superior to another. We also had concerns about the motherboard. Having integrated graphics is convenient for testing purposes, but its VRMs get very hot when placed on load for long periods, and we have to add extra cooling on occasion to avoid testing irregularities which sometimes crop up.

We experimented extensively with a AMD Phenom-II 965 CPU that has a TDP of
125W, with overclocking and overvolting the Pentium D 950 to make it run hotter
(over 100W at the AUX12V socket), and with a 130W TDP Intel Core 2 Extreme QX9650
CPU. Many different motherboards were tried during the selection process.

Ultimately we decided to upgrade to a Core i7 setup using the 3.2GHz i7-965
Extreme and an Asus P6T SE motherboard. The i7-965 is one of
the fastest desktop processors with a 130W TDP, so it should get heatsinks nice
and toasty. The P6T SE has two sets of mounting holes, for both
LGA1366 and LGA775 coolers. This means we can use the same platform for LGA1366
heatsinks as well as those with only LGA775 compatibility. The board also has
large heatsinks for the VRMs and the NB chip, which bodes well for stability
and longevity under high stress.

Surprisingly, the 130W TDP Intel QX9650 drew only 66W at peak load (including
losses in the VRM) on the existing Asus P5Q-EM platform — it ran 20W
cooler than the old Pentium D 950. (This result was closely matched
in a more elaborate test by Lost
Circuits.) A Phenom II/AM3 setup was also considered, but rejected in
the end due to the extra wear we’d put on the board since the stock retention
bracket would have to be removed/reattached often.

i7 represent a maximum of perhaps 10% of Intel’s processor sales at this time,
a small minority. The i7 processors have been widely recognized as not only
the most powerful CPUs, but also the most power hungry. By choosing this processor
as the basis of our test platform, we’re automatically more focused on cooling
very hot processors quietly. This is a tacit recognition that today’s typical
mainstream processors are no longer a challenge to cool quietly.

Methodology Changes?

We also asked ourselves whether this environment was truly optimal for heatsink
testing, after all most systems have the motherboard mounted vertically with
the heatsink protruding outward to the side. However, mounting the board in
an actual case would increase the wear and tear on the board dramatically, especially
since it already is subject to potential damage each time we install a CPU cooler;
the methods manufacturers have invented to mount their heatsinks can be scary
at times. While replacing a broken board is trivial, SPCR is not an entity with
unlimited resources.

It has also been argued that the platform should be enclosed to simulate in-system conditions. If so, other considerations have to be made. What type of enclosure? One with the power supply at the bottom or top? Where should the fan placements and ventilation points be located and what size? Should we add a system fan, and if so, which one, and set at what speed? Going this route may lead to realistic results, but for only one case layout. Also, any extra variables we introduce would benefit some heatsinks while being a detriment to others. For a fair test of heatsink efficiency, we need to eliminate as many performance-affecting variables as possible. For isolating the heatsink, our original, open system setup seems optimal.

Our only change in methodology will be to test high performance heatsinks both with stock CPU settings and overclocked/overvolted. The latter will push the chip’s thermal envelope, helping to separate the best coolers from one another.

UPDATED CPU HEATSINK TEST PLATFORM

Key Components

• Intel Core i7-965 Extreme
  Nehalem core, LGA1366, 3.2GHz, 45nm, 130W TDP.
• Asus
  P6T SE ATX motherboard. A LGA1366 X58 chipset board with short solid-state
  capacitors around the CPU socket, low profile northbridge and VRM heatsinks,
  and mounting holes for both LGA1366 and LGA775 coolers.
• Asus
  EAH3450 Silent graphics card.
• Intel
  X25-M 80GB 2.5″ solid-state drive. Chosen for silence.
• 2GB QiMonda
  DDR3 memory. 2 x 1GB DDR3-1066.
• Seasonic X-650 SS-650KM
  650W ATX power supply. This PSU is semi-passively cooled. At the power levels
  of our test platform, its fan does not spin.
• Arctic Silver
  Lumière: Special fast-curing thermal interface material, designed
  specifically for test labs.
• Nexus 120 fan (part of our standard testing methodology; used when
  possible with heatsinks that fit 120x25mm fans)

The system is silent under the test conditions, except for the fan on the heatsink, which is a controlled variable.

The Asus P6T SE has short but effective northbridge and VRM heatsinks, allowing for plenty of clearance around the CPU.

Measurement and Analysis Tools

• Extech 380803 AC power analyzer/datalogger for measuring AC power
  at the wall to ensure that the heat output remains consistent.
• Custom-built, four-channel variable DC power supply, used to regulate
  the fan speed during the test.
• PC-based spectrum analyzer:
  SpectraPlus with ACO Pacific mic and M-Audio digital
  audio interfaces.
• Anechoic chamber
  with ambient level of 11 dBA or lower
• Various other tools for testing fans, as documented in our
  standard fan testing methodology.
• SpeedFan,
  used to monitor the on-chip thermal sensors. The sensors are not calibrated,
  so results are not universally applicable. The hottest core reading is used.
• Prime95,
  used to stress the CPU heavily, generating more heat than most real applications.
  8 instances are used to ensure that all 4 cores (with Hyper-threading) are
  stressed.
• CPU-Z,
  used to monitor the CPU speed to determine when overheating occurs; throttling
  has been observed to occur at between 95~100°C.

Testing Methodology

As in the past, the main question we ask in our review is, What is the cooling
power of this heatsink with this quiet fan whose characteristics are well known?
By asking this question, we put all the heatsinks on the same playing field
— no screaming 100 CFM fans. All have only the aid of the same quiet, low
airflow fan. The heatsink, then, is the only unknown variable. This approach
guarantees that all heatsinks are tested under the same acoustic and airflow
conditions.

When a standard fan is included, we run two sets of tests, one with our reference
fan, and one with supplied fan. Stock fans are profiled according to our
standard fan testing methodology, which uses a similar noise-centric approach.
With heatsinks that have a fully integrated fan not easily replaced, it is tested
as delivered.

Reference Nexus 120mm fan measurements
Voltage	SPL@1m	Speed
12V	16 dBA@1m	1100 RPM
9V	13 dBA@1m	890 RPM
7V	12 dBA@1m	720 RPM
5V	11 dBA@1m	530 RPM

Load testing is accomplished using Prime95 to stress the processor, and the
graph function in SpeedFan is used to make sure that the load temperature is
stable for at least ten minutes. Larger, high performance heatsinks will undergo
an additional test with the CPU overclocked to 3.6GHz and overvolted to 1.40V.
The stock fan is tested at various voltages to represent a good cross-section
of its airflow and noise performance.

We assess the heatsink and mounting mechanism together as a unit. A
heatsink’s intrinsic cooling power is determined mainly by:

• its radiating surface area
• the heat transfer coefficient of its materials
• the spacing and number of fins
• its geometry
• the smoothness and flatness of the CPU contact surface
• overall mass
• ease and efficacy of the mounting mechanism

The mounting mechanism is mentioned because it maintains the all-important
contact between CPU and heatsink. The amount of pressure brought to bear on
the interface also affects cooling. It is also the only real interface
between HS and user. We may say we use a HS, but it’s not the same way
that we use a car, for example. We interact constantly with a car while
using it. User interaction with a HS really happens only when the HS is installed
or uninstalled. If this design aspect is poor and results in the user having

difficulty with installation, or failing to mount the HS correctly, then poor
cooling of the CPU can result. Some mounting mechanisms are poor, both difficult
to install and lacking in precision or security; others are integrated wonderfully
into the heatsink and easy to use. The mounting mechanism is a critical part
of the HS design.

Test platform.

Power consumption is also monitored, both from the wall and from the
ATX12V connector. An 0.01 ohm shunt resistor is placed in-line and the voltage
drop (in mV) is measured with a high precision digital multimeter. Knowing the
resistance and the voltage across the resistor gives us enough information to
calculate the current using Ohm’s Law. A second multimeter measures the voltage
going into the connector which is around 12V. Multiplying the current by the
voltage gives us the power draw of the CPU and VRMs in watts. By monitoring
power consumption at this point, we can possibly spot increases in VRM efficiency
caused by airflow generated by top-down coolers.

Measuring the voltage across the shunt resistor.

CPU Power Consumption

Test Platform Power Consumption (DC)
Platform	System	AUX12V
Pentium D 950 + Asus P5Q-EM	111W	86W
Core i7-965 + Asus P6T SE + Radeon HD 3450	148W	86W

Now you may be wondering exactly how much more power the i7-965 uses compared
to the Pentium D 950. Unfortunately, that is a rather difficult question to
answer. With socket 775 processors, all the power for the CPU channels through
the AUX12V connector. This is not true for socket 1366 processors.

We measured the power consumption at the AUX12V plug to be about the same on
both platforms, 86W including losses in the VRMs. The i7’s system power was
37W higher and it definitely put out considerably more heat as some of the heatsinks
we re-tested failed at lower fan speeds, causing the CPU to throttle down. We
had calculated the HD 3450 graphics card power consumption at idle to be 11W.
The boards themselves use the about same amount, so there is some 26W unaccounted
for.

Lost
Circuits has a detailed article about the power configuration of the
i7. As it turns out, the i7 CPU also draws power from the +5V and +3.3V rails
through the motherboard’s main 24-pin power connector for the integrated memory
controller. The author states that Intel’s own specifications indicate the total
non-AUX12V power to the i7 “can weigh in with as much as 44.58 Watts.”
This accounts for the extra power we measured for the i7 system, and leaves
us without an accurate power consumption figure for the i7-965 CPU.

We can’t say with certainty that the unaccounted 26W is the extra power going
into the i7-965 via the +5V and +3.3V lines, but if it is, then the CPU power
would be 112W, which is a bit closer to the rated 130W TDP than the 86W we measured
at the AUX12V socket. In any case, since we cannot determine the i7-965’s power
draw accurately, our traditional degrees Celsius rise per watt of CPU power
(°C/W) will have to be shelved; instead we will rely solely on temperature
rise over ambient.

Some readers may note the PSU change from the old platform: A new Seasonic
X-650 80 PLUS Gold rated power supply instead of the previous Silentmaxx
Fanless 400W MX460-PFL01. There were two main reasons for the change:

1. The X-650 PSU is far more efficient, drawing only 168W AC under load with
   the new test platform with default BIOS settings, compared to the Silentmaxx’s
   180W AC. We like to minimize our energy consumption whenever possible.
2. The X-650 actually runs much quieter; it is essentially silent under the
   test load. In contrast, under the load of the new i7 platform, the Silentmaxx
   made enough electronic noise (buzzing, humming — well over 20 dBA@1m)
   that its fanless operation became moot.

For those who are interested, here’s a snapshot of the power profile of our
new platform, cooled with a Prolimatech Megahalems and the reference fan at
12V. The power consumption can vary substantially with different heatsinks,
especially with low airflow, due to the VRMs becoming less efficient when they
get really hot. Just before throttling occurs, AC power can be some 20W higher
and the AUX12V power can be 16~18W higher.

i7-965 Platform Power Consumption at Load (w/ Prolimatech Megahalems heatsink & reference 12cm fan @1000 rpm)
settings	System (AC)	System (DC)	AUX12V (DC)	CPU (guesstimate)
Default BIOS	168W	148W	86W	112W
CPU at 3.6 GHz, 1.4V	215W	194W	130W	160W

 

TEST RESULTS

To inaugurate our new heatsink platform and to establish a baseline of new
data, we put eight of the top coolers tested previously through the paces once
again. The eight coolers in question are, in alphabetical order, the Noctua
NH-C12P and NH-U12P,
the Prolimatech Megahalems, the Scythe
Kabuto and Mugen-2,
the Thermalright Ultra-120
and Ultra-120 eXtreme, and the Zalman
CNPS10X Extreme.

Top contenders.

Old Results

First let’s take a look at the results from our previous test platform.

°C rise Comparison: Pentium D 950 (Asus P5Q-EM)
Heatsink	Nexus 120mm fan voltage / SPL @1m
	12V	9V	7V
	16 dBA	13 dBA	12 dBA
Prolimatech Megahalems	10	14	17
Thermalright Ultra-120 eXtreme	12	14	17
Scythe Kabuto	13	15	19
Noctua NH-U12P	14	16	17
Zalman CNPS10X Extreme	14	17	21
Scythe Mugen-2	15	17	19
Thermalright Ultra-120	15	17	21
Noctua NH-C12P	16	18	21

As you can see, the Pentium D 950 gave us a rather narrow range of performance
differences among the eight heatsinks. Many of the heatsinks on the chart are
only 1~2°C warmer or cooler than its closest rival, especially at the fan
voltages of 12V and 9V. Only 4~6°C separates the 1st and 8th place at each
fan setting. A hotter CPU would definitely help differentiate this huddled mass.

Updated Results: Stock Speed

°C rise Comparison: i7-965 @ Stock (Asus P6T SE)
Heatsink	Nexus 120mm fan voltage / SPL @1m			i7-965 Rank	Pentium D 950 Rank
	12V	9V	7V
	16 dBA	13 dBA	12 dBA
Prolimatech Megahalems	35	39	42	#1	#1
Scythe Mugen-2	37	40	43	#2	#5
Noctua NH-U12P	38	40	41	#2	#3
Thermalright Ultra-120 eXtreme	38	41	45	#4	#2
Zalman CNPS10X Extreme	39	43	48	#5	#5
Thermalright Ultra-120	42	44	49	#6	#7
Scythe Kabuto	43	48	54	#7	#3
Noctua NH-C12P	44	47	54	#7	#8

On our i7-965 platform, a somewhat different picture develops, with 1st and
8th separated by 9~13°C depending on the fan speed. The Megahalems stays
on top, while the Scythe Mugen-2 and Noctua NH-U12P trail it only slightly,
locked in a virtual tie. The Noctua is particularly impressive, performing only
3°C worse with the fan at 7V compared to 12V. The Ultra-120 eXtreme comes
in 4th due to its slightly higher temperature at 7V; its tight fin spacing hurts
it here. The downblowing Kabuto and NH-C12P tie for last place, with particularly
poor performance at 7V compared to the competition.

Updated Results: Overclocked

°C rise Comparison: i7-965 @ 3.6GHz, 1.40V (Asus P6T SE)
Heatsink	Nexus 120mm fan voltage / SPL @1m			i7-965 OC Rank	i7-965 Stock Rank
	12V	9V	7V
	16 dBA	13 dBA	12 dBA
Prolimatech Megahalems	50	53	59	#1	#1
Thermalright Ultra-120 eXtreme	51	56	63	#2	#4
Noctua NH-U12P	53	56	60	#2	#2
Scythe Mugen-2	53	56	62	#4	#2
Thermalright Ultra-120	56	60	67	#5	#6
Zalman CNPS10X Extreme	60	66	FAIL	#6	#5
Scythe Kabuto	62	70	FAIL	#7	#7
Noctua NH-C12P	64	73	FAIL	#8	#7

When the i7-965 is overclocked to 3.6GHz and overvolted to 1.40V, the difference
between 1st and 8th expands from 8°C to 20°C at 9V. At 7V, some of the
heatsinks aren’t able to keep the processor cool enough, causing the CPU to
throttle. This happens when any of the cores reaches 95~100°C (typically
75~80°C thermal rise). The CNPS10X Extreme and the two top-down coolers,
the Kabuto and NH-C12P, couldn’t handle the heat with the reference fan at 7V.

The Megahalems continues to dominate, with an even bigger lead over the #2
spot shared by the Ultra-120 extreme and NH-U12P. The Noctua gets the best of
the U120E at 7V, while the U120E takes the lead at 12V. The race is fairly tight
with the Mugen-2 just nipping at the NH-U12P’s heels. The vanilla Ultra-120
is about 4°C worse on average at the #5 spot. The CNPS10X, Kabuto and NH-C12P
are all significantly poorer, running hotter by double-digits compared to the
Megahalems, and failing at the 7V level.

As an aside, it’s no surprise that the strongest performers have sturdy, spring-loaded,
bolt-through mounting. The Megahalem’s superior mounting system (detailed in
the original review) is certainly an important part of its excellent and consistent
performance.

MP3 SOUND RECORDINGS

These recordings were made with a high
resolution, lab quality, digital recording system inside SPCR’s
own 11 dBA ambient anechoic chamber, then converted to LAME 128kbps
encoded MP3s. We’ve listened long and hard to ensure there is no audible degradation
from the original WAV files to these MP3s. They represent a quick snapshot of
what we heard during the review.

These recordings are intended to give you an idea of how the product sounds
in actual use — one meter is a reasonable typical distance between a computer
or computer component and your ear. The recording contains stretches of ambient
noise that you can use to judge the relative loudness of the subject. Be aware
that very quiet subjects may not be audible — if we couldn’t hear it from
one meter, chances are we couldn’t record it either!

The recording starts with 10 second segments of room ambiance, then the fan
at various levels. For the most realistic results, set the volume so that
the starting ambient level is just barely audible, then don’t change the volume
setting again.

• Nexus
  120mm Real Silent Case fan at one meter
  — 5V (11 dBA@1m)
  — 7V (12 dBA@1m)
  — 9V (13 dBA@1m)
  — 12V (16 dBA@1m)

FINAL THOUGHTS & RANKINGS

Changing our platform to a more power hungry setup has revealed the weaknesses
of top-down coolers. It hasn’t however, created any new winners. The heatsinks
that excelled today also did so on our old test platform, though the increased
range of results has created more separation, making it easier to proclaim one
cooler better/worse than another. We look forward to building on these results
throughout the following months as we have several interesting and intimidating
heatsinks in the lab awaiting testing.

#1 Prolimatech Megahalems
(Street Price: $60~$65 without fan):

There is very little to say about the Megahalems and we mean that in the best
possible way. It topped the charts of our old test platform and remains unbeaten
on our new platform. It’s a large heavy cooler with an excellent mounting system
and can be found for $60. That’s a reasonable amount for a topnotch heatsink.

#2 Thermalright Ultra-120
extreme (Street Price: $55 without fan, $70 with fan):

The Ultra-120 eXtreme was once a champion at SPCR, but has since lost its top
billing to the Megahalems. However, placing a close second is hardly shameful
considering it has been two and a half years since its release. It has a simple
and secure mounting scheme and is more widely available than the Megahalems.
Its only fault is its slightly poorer performance than other heatsinks in its
class when the fan speed is dialed back to very low levels. LGA1366-compatible
models can be purchased for as low as $55 at various e-tailers.

#2 Noctua NH-U12P
(Street Price: $55~65 with fan):

The NH-U12P is a classic heatsink that has withstood the test of time very
well. It has a solid backplate mount and is an extremely versatile cooler performing
well with both high and low airflow. On our overclocked test platform, the temperature
only rose 7°C when the fan was undervolted from 12V to 7V, the smallest
difference of the entire field. The stock fan does sound a bit annoying at full
speed, but it undervolts well and conveniently ships with a 7V and 9V in-line
3-pin adapter. It can be found for $55 online which is a very good deal if you
decide to keep the fan.

#4 Scythe Mugen-2
(Street Price: $35~$40 with fan):

The Mugen-2 lost to the NH-U12P by a hair, only being edged out by a slim margin
with our reference fan at low speed. It does suffer from one major flaw though:
installation is very troublesome with the final mating between the backplate
and heatsink requiring the motherboard to be flipped upside down. Of the 8 coolers
in this roundup, the Mugen-2 is the only one that requires motherboard removal
to unmount, even temporarily for a processor upgrade or a fresh application
of thermal compound. This inconvenience may be worth it though, as it is an
extremely proficient cooler, ships with a very quiet fan, and can be acquired
as little as $35. It is far and away the best value of the bunch.

#5 Thermalright Ultra-120
(Street Price: N/A, discontinued?):

The smaller brother of the Ultra-120 extreme performed fairly well, a few degrees
off from elite status. The 6-heatpipe U120E posted results 5°C better than
the 4-heatpipe U120.

#6 Zalman CNPS10X Extreme
(Street Price: $70~$80 with fan, $60~$65 for the plain ‘Quiet’ version):

The CNPS10X Extreme unfortunately does not perform like a $70 heatsink. It
trailed the Ultra-120 by about 5°C with our reference fan at 12V and 9V,
and at 7V it failed to prevent our overclocked i7 from throttling. Performance
and price isn’t its only weaknesses either. The mounting method for LGA1366
does not use a backplate. Instead, a plastic retention frame is screwed on and
the heatsink mounts to that. And while the stock fan sounds fairly decent when
undervolted, it cannot be replaced without modification. Zalman also sells the
CNPS10X Quiet which is basically the same cooler but with standard fan mounting
and without the Extreme’s fancy fan controller. At $60, it’s a better deal,
but still a questionable value when you consider the alternatives.

#7 Scythe Kabuto
(Street Price: $45~$50 with fan):

On the Pentium D test platform, the Kabuto was the best top-down cooler tested,
and rivaled the elite tower heatsinks. On an i7, it can do the job, but with
far less proficiency. When we overclocked the i7, the Kabuto came within 15°C
of CPU throttling with our reference fan running at 12V and couldn’t prevent
it from happening at 7V. Like the Mugen-2, it comes with a very nice fan, but
like older Scythes, it uses pushpins to mount. For $45, we’d rather buy the
Mugen-2 and pocket the difference.

#8 Noctua NH-C12P
(Street Price: $65~$70 with fan):

The NH-C12P was edged out by the Kabuto by only a couple of degrees, but they
were both equally poor compared to the tower heatsinks tested. Blowing hot exhaust
air out to the side is far more efficient than pushing it down toward the PCB,
especially when the CPU temperature gets very high. We should also note that
our power consumption measurements did not change when when using either down-blower,
so the extra airflow did not improve VRM efficiency. On a board like the Asus
P6T SE, with ample VRM cooling, it would appear that these types of heatsinks
have little practical value. If that wasn’t enough of a deterrent, there`s also
the $65 price-tag.

A cautionary note should be sounded regarding the results from our new
heatsink test rig: The differences that show up in our testing will not be as
relevant if your CPU is a typical sub-100W TDP model. This is not to say the
8th ranked heatsink will perform as well as the top ranked heatsink if you use
it with a 65W TDP processor. But while the 8th ranked heatsink will allow an
overclocked i7 to overheat and throttle if you run it with a fan under 700rpm,
it will cool a 65W processor perfectly well. In essence, the new test rig is
most useful for those who are running hot CPUs. Depending on reader response,
we may add an undervolted CPU test run to make the reviews 100%
relevant no matter how hot or cool a CPU you run.

FLASH! See important 10 April 2010 Postscript overleaf.

* * *

Articles of Related Interest
CPU Heatsink Test Platform, 2006-2009

ZEROtherm Nirvana CPU Cooler
Smallish LGA775 Heatsink Roundup
– Part 2
Scythe Top-Down Coolers:
Kabuto vs. Zipang 2
LGA775 Low Profile Heatsink
Roundup
Scythe Mugen-2 CPU Cooler
Scythe Katana 3: Same slant, new version

* * *

Discuss
this article in the SPCR forums.

POSTSCRIPT: O THE TESTING LIFE! April 10, 2010 by Lawrence Lee & Mike Chin

Two months after this article was published, while using the i7 test platform
after a several weeks hiatus to test a new pair of heatsinks, we noticed that
the temperature readings had changed… and perhaps become unstable. The heatsinks
in hand were the Cogage TrueSpirit, a cut-down version of the Thermalright Ultra-120,
and the Zalman CNPS10X Quiet, which has wider fin spacing than the CNPS10X Flex
and Extreme variants. The Ultra-120 and CNPS10X Extreme had finished in 6~7th
position on our chart in February
(when cooled with the reference 120mm fan), so we were stunned to see the TrueSpirit
and CNPS10X Quiet both performing at levels rivaling the Prolimatech Megahalems
and Noctua NH-D14, the current champions. The tests were repeated twice, with
the same results.

This called for a complete retest of several reference heatsinks, at the end
of which we were dismayed to discover that the core temperature in the test
platform had dropped significantly, basically overnight, and every heatsink
was suddenly producing cooler CPU temperatures. In addition, we noticed that
both temperature and power consumption was swinging high/low at random intervals
after a period of CPU stress operation, but only when the system was overclocked/overvolted.

After much more experimentation, we learned that the same behavior was evident
with different processors, so we suspected the board had been damaged, perhaps
by the 1.4V overvolt applied in our stress testing. In the end, we had no choice
but to replace the board. We asked Asus for a new motherboard with better power
regulation, and in response, they sent us the current top-of-the-line LGA1366
board, the P6X58D, which uses a 16+2 phase power design.

Asus sent us one of their new top-of-the-line P6X58D Premium to be our replacement i7 heatsinks test platform. It has the benefit of a sophisticated 16 + 2 phase voltage regulation module design.

After some testing on the new board, we found that the problem persisted,
though it did take longer for the temperature and power to begin fluctuating.
Dialing back the CPU voltage helped somewhat. After a few more days of systematic
testing, we decided the system should be run at stock settings from now on,
to avoid any potential inconsistentcies in the power/thermal load. We would
prefer to have the processor run as hot as possible, but it is more important
that we can obtain stable results over a lengthy period of time, due to the
method in which we conduct our heatsink testing (typically with both the stock
and reference fans at various speeds).

i7-965 Platform Power Consumption at Load (w/ Prolimatech Megahalems heatsink & reference 12cm fan @1000 rpm)
Motherboard (RAM configuration)	Setting	System (AC)	System (DC)	Thermal Rise
P6T SE (single channel)	Stock	168W	148W	35°C
	3.6 GHz, 1.4V	215W	194W	50°C
P6X58D Premium (triple channel)	Stock	199W	179W	38°C

One of our readers had noted that we were testing using only a single channel
memory configuration and that more memory would increase power draw, so we
switched to triple channel RAM. This change had an enormous effect on power
consumption, bumping it up by more than 30W at the AC outplet. Oddly, however,
the increase CPU temperature was virtually nil. There was also little increase
in the current at the AUX12V socket, which suggests that the memory controller
was responsible for much of the power increase, on the 5V and 3.3V lines.

In any case, the change in motherboard and test platform settings meant a complete
retest of all the heatsinks we’ve tested on the i7 platform thus far. This was
a tedious, repetitive task that took two days, even though only the Nexus 120
fan was used.

Heatsink Re-test Results

Original i7 Test Results: °C rise Comparison (Stock Settings)
Heatsink	Nexus 120mm fan voltage / SPL @1m			Rank
	12V	9V	7V
	16 dBA	13 dBA	12 dBA
Noctua NH-D14	36	38	41	#1
Prolimatech Megahalems	35	39	42	#1
Scythe Mugen-2	37	40	43	#3
Noctua NH-U12P	38	40	41	#3
Thermalright U120 eXtreme	38	41	45	#5
Zalman CNPS10X Extreme	39	43	48	#6
Thermalright U120	42	44	49	#7
Zalman CNPS10X Flex	42	45	49	#7
Scythe Kabuto	43	48	54	#9
Noctua NH-C12P	44	47	54	#9

There were only a few changes in rank. The CNPS10X Extreme slipped from #6
to #8 due to an average 4°C increase in thermal rise. The downblowing Kabuto
fared 6°C worse, remaining in last place but this time alone as the NH-C12P
held its ground and actually improved two ranks because the Zalman CNPS10X Extreme
and Flex slipped slightly. At the top of the spectrum, little changed with the
Megahalems and NH-D14 battling for top position, though this time joined by
the smaller NH-U12P.

New i7 Test Results: °C rise Comparison
Heatsink	Nexus 120mm fan voltage / SPL @1m
	12V	9V	7V	Rank
	16 dBA	13 dBA	12 dBA
Prolimatech Megahalems	38	41	44	#1
Noctua NH-D14	38	42	45	#2
Noctua NH-U12P	39	42	44	#2
Scythe Mugen-2	39	42	45	#4
Thermalright U120 eXtreme	40	43	48	#5
Thermalright U120	42	45	49	#6
Noctua NH-C12P	43	47	51	#7
Zalman CNPS10X Extreme	43	47	53	#8
Zalman CNPS10X Flex	45	50	54	#9
Scythe Kabuto	51	53	60	#10

The new results are indicative of at least one significant aspect of heatsink
testing: It seems almost impossible to get exactly the same results from one
set of tests to the next (even when the same CPU and motherboard are used).
There are a couple of possible reasons:

• The thermal measurement tools simply don’t have fine enough resolution;
  perhaps the accuracy of the digital termistor in the CPU is worse than +,-1°C.
• It may be impossible to apply the same amount of TIM and get exactly the
  same degree of tension on the heatsink during mounting, the end result being
  variations in results from different instances of testing the same heatsink
  on the same platform.

For now, the new i7 test results table directly above will remain the reference
point for comparing against future heatsink tests. We move on.

* * *

FLASH! See 31 May 2010 Postscript on Test Platform for
Smaller Heatsinks, overleaf.

Discuss
this article in the SPCR forums.

POSTSCRIPT 2: 31 May 2010 A TEST PLATFORM FOR SMALLER HEATSINKS

It was always our intention to bring back the old socket 775 Pentium D950 platform
to review heatsinks that are more suitable for midrange and lower power CPUs.
This would be done when such heatsinks came our way. The impetus came in the
form of Gelid Silent Spirit and Scythe Samurai ZZ heatsink samples.

We’ve already mentioned issues with keeping the VRMs on the old 775 test platform
adequately cooled. So two other Intel chipset socket 775 boards were explored;
both overclocking oritented, with massive heatsinks on the VRMs and the Northbridge.
A week of experimentation with these board led only to frustration. The temperature
monitoring on both boards turned out to be unreliable and unstable. The boards
themselves were basically fine, but under long CPU/thermal stress testing, the
monitoring chips seemed to go haywire, giving unreliable and inconsistent temperature
readings.

At the end of the week, we decided to try a new route: An AMD AM3 CPU and AMD
785 chipset motherboard, the latter with large heatsinks for both VRMs and Northbridge
chip. A few days of experimentation were enough to establish that this combination
was stable, had the appropriate thermal load for our requirements, and provided
consistent temperature monitoring. It has become our test platform for smaller
and low profile heatsinks. The details are as follows:

Key Components in Smaller Heatsink Test Platform:

• AMD Athlon II X4 630 AM3,
  2.8GHz, 45nm, 95W TDP.
• Asus M4A785TD-V EVO ATX motherboard.
  785G chipset.
• Kingston
  SSDNow V 30GB 2.5″ solid-state drive. Chosen for silence.
• 2GB
  Corsair Dominator DDR3 memory. 2 x 1GB DDR3-1800 in dual channel.
• FSP Zen 300W
  ATX power supply. Fanless.
• Arctic Silver
  Lumière: Special fast-curing thermal interface material, designed
  specifically for test labs.
• Nexus 92 fan (part of our standard testing methodology; used when
  possible with heatsinks that fit 92x25mm fans)

Asus M4A78TD-V EVO board on our usual 2-tier open platform, with fanless PSU and SSD on lower level, and shunt resistor on AUX12V connector to monitor CPU power.

With a fanless power supply and a solid state drive, the test system is silent
under the test conditions, except for the CPU cooling fan(s). At full load,
the total system power draw is 132~140W AC, with the CPU and VRMs drawing 85~91W
DC (measured at the AUX12V connector), depending on their respective temperatures.

Smaller Heatsink Test Platform: Full Load Power Details
System	132-140W AC
CPU+VRM	85~91W DC

Normally, our reference fan is used whenever possible, the measured details
of which are shown below.

Reference Nexus 92 mm fan Anechoic chamber measurements
Voltage	SPL@1m	Speed
12V	16 dBA	1470 RPM
9V	12 dBA	1150 RPM

Measurement and Analysis Tools

• Extech 380803 AC power analyzer / data logger for measuring AC system
  power.
• Custom-built, four-channel variable DC power supply, used to regulate
  the fan speed during the test.
• PC-based spectrum analyzer:
  SpectraPlus with ACO Pacific mic and M-Audio digital
  audio interfaces.
• Anechoic chamber
  with ambient level of 11 dBA or lower
• Various other tools for testing fans, as documented in our
  standard fan testing methodology.
• SpeedFan,
  used to monitor the on-chip thermal sensors. The sensors are not calibrated,
  so results are not universally applicable.
• CPUBurn,
  used to stress the CPU heavily, generating more heat than most real applications.
• CPU-Z,
  used to monitor the CPU speed to determine when overheating occurs.
• Thermometers to measure the air temperature around the test platform
  and near the intake of the heatsink fan.

Noise measurements are made with the fans powered from the lab’s variable DC
power supply while the rest of the system was off to ensure that system noise
did not skew the measurements.

CPUBurn is used to stress the processor, and the graph function in SpeedFan
used to ensure that the load temperature is stable for at least ten minutes.
The stock fan is tested at various voltages to represent a good cross-section
of airflow and noise performance.

A few of the smaller heatsinks tested on the socket 775 system were retested
on the new AM3 platform to establish some reference points. The ambient conditions
during testing were 10~11 dBA and 21~23°C.

Reference Heatsink Performance:

Scythe Ninja Mini w/ ref. 92 mm fan
Fan Voltage	SPL@1m	Temp	°C Rise
12V	16 dBA	46°C	23
9V	12 dBA	50°C	27
°C Rise: Temperature rise above ambient (23°C) at load.

 

Arctic Alpine 64 w/ stock fan
Fan Voltage	SPL@1m	Temp	°C Rise
12V	28 dBA	45°C	22
9V	23 dBA	52°C	29
7V	17 dBA	57°C	34
6V	15 dBA	66°C	43
5V	12 dBA	69°C	46
°C Rise: Temperature rise above ambient (23°C) at load.

 

Xigmatek HDT-SD964 w/ stock fan
Fan Voltage	SPL@1m	Temp	°C Rise
12V	34~35 dBA	40°C	17
9V	26 dBA	41°C	18
7V	15 dBA	45°C	22
6V	13 dBA	50°C	27
5V	11~12 dBA	57°C	34
Xigmatek HDT-SD964 w/ ref. 92 mm fan
12V	16 dBA	47°C	24
9V	12 dBA	53 °C	30
°C Rise: Temperature rise above ambient (23°C) at load.

 

Scythe Big Shuriken w/ stock fan
Fan Voltage	SPL@1m	Temp	°C Rise
12V	28 dBA	44°C	22
10V	24 dBA	47°C	25
9V	20 dBA	48°C	26
8V	16 dBA	52°C	30
7.3V	11 dBA	59°C	37
Scythe Big Shuriken w/ ref. 120 mm fan
12V	16 dBA	46°C	24
9V	13 dBA	50°C	28
7V	12 dBA	55°C	33
°C Rise: Temperature rise above ambient (22°C) at load.

The new AM3 setup rounds out our CPU heatsink testing — at least for the
forseeable future. The two test platforms should provide silence-oriented PC
enthusiasts a wealth of information to choose an appropriate CPU heatsink for
their requirements, whether it’s for a hot cutting-edge system or a modest middle-of-the-road
all arounder.

* * *

Discuss
this article in the SPCR forums.