You are here

Review: Thermalright SP94, SP97 & other heavyweights

Dec 22, 2003 by Mike
Chin

Products & approx. market prices in US$

Thermalright SP94 P4 Heatsink $50

Thermalright SLK900U on P4/socket A $40

Thermalright SP97 socket-A heatsink $50

Zalman ZM7000A CU/AlCu on P4/socket A
$35~50

Suppliers

Thermalright



Sharka Corp

Silicon Valley Compucycle (SVC)

Thermalright has held a lofty position at the top of the heatsink ranks here among silence seekers. They are also regarded as tops by legions of overclockers, gamers and performance enthusiasts elsewhere. They provide cooling performance second to none when you're using a banshee fan approaching 100 CFM yet do amazingly well even when you are babying your hearing with a gently undervolted Panaflo.

Thermalright's ascendance goes back at least to their SLK800, introduced over 18 months ago. The SLK900 picked up and continued where the SLK800 left. But earlier this year, the ZM7000 flower-turned-radial Zalman heatsink stepped into the fray, and since then the race has been very close.

The SP94 is Thermalright's latest heavy metal flagship for the P4, introduced a few short months ago. It incorporates three heatpipes in an all-copper structure that is a natural evolution of their previous products. The just-introduced SP97 is Thermalright's socket-A version of the SP94. These heatsinks represent the best Thermalright has to offer at this in time. The progress is more evolutionary than revolutionary -- at a quick glance, they look much like the SLK900 with a bit of additional plumbing. (However, close examination shows other differences; more on this later.)

The other heatsinks are listed at the top of the page because even though this review is ostensibly about the SP97 and SP94, the other heatsinks were needed as reference points for comparison. Furthermore, the SLK900U had not been tested with an AMD CPU before, and neither had either of the Zalman ZM7000s, which in the most recent A version, can be used with socket-A. These heatsinks, already reviewed in P4 applications, get coverage here in the context of socket-A CPU cooling. On top of all this, a change was made to the P4 test platform: The P4-1.8 was retired (too cool) and a much hotter P4-2.8 introduced. Finally, my socket-A platform was upgraded from the now ancient KT-133A to a much more recent Asus and a XP2500+ with Barton core.

As you might guess, this is going to be a wild and wholly heatsink roundup. With the profusion of models that work with both P4 and Socket-A, it's hard to keep it all neat and tidy. And we have not even touched the Athlon 64 yet! So hang on to your... er, fans?... and bear with me please.

SP94 & 97 Specifications

SP94
SP97
All copper design for maximum performance

Soldered fins to base to ensure effective contact

92mm stock fan for more cooling capability

Screws through motherboard mounting for LAN party

L99 x W59 x H45 (mm) Top, without fan

L67 x W81 x H45 (mm) Bottom, without fan

590 grams (heatsink only)

L99 x W59 x H50 (mm) Top, without fan

L75 x W55 x H50 (mm) Bottom, without fan

585 grams (heatsink only)

Details

As the top photo shows, the SP94 and SP97 look very similar. In terms of their main fins structure, they are virtually identical in size to the SLK900. There are a few key differences between the SP94 and SP97, mostly related to the requirements of the difference CPU form factors they are meant for.

The first of these is the base and mounting structure. The SP94, on the right in the photo above, has 4 hefty square nubs that extend out of the corners of base. Each of these has a hole in it. These are the mounting holes. They line up with the holes around the CPU socket in any P4 motherboard -- once the usual plastic heatsink retention bracket is removed.

The SP97, on the left above, also has 4 lugs with holes in them, but the lugs are not quite as hefty, and they line up with the 4 HS mounting holes that can be found on some AMD socket-A motherboards. These through-the-board HS mounting holes are not nearly as common as they used to be. Some months ago, AMD announced the cessation of official support for this HS mounting scheme. (I don't believe this was ever a requirement for AMD socket-A boards. It is hard to imagine the benefit of removing the recommendation or guideline, however. It certainly makes life harder for performance HS companies that have relied on these mounting holes -- which undoubtedly allow the secure use of far heavier heatsinks, especially in transit. One imagines some backroom politicking behinds the scenes at AMD.)

Two other minor differences exist:

  • The SP94 base is very slightly thinner (though bigger in area than the SP97 base), making the SP97 slightly taller.
  • The SP94 has one more fin than the SP97.

One significant design departure for the SP94/97 is that the base structure is flat, and the fins are bonded directly to the base. The SLK800 and 900 both utilize a heavy triangular-cross section tunnel running the length of the heatsink. This is visible in photos in the SLK900 review as well as the SLK800 photo below.



SLK800's triangular-cross section "tunnel": This design feature is not used in the SP94/97.

The photo below show the three heatpipes, fine fins and a base that's flat on both sides. One might conjecture that the triangular-cross section tunnel, being a part of the base and connected quite high up on the fins, has an effect similar to the heatpipes.

The weight of the SP94 and SP97 is essentially the same, at 590 and 585 grams without a fan. That's quite substantial, and a standard 80x25mm fan will typically add about 100 grams, and more if you opt for a larger 92mm fan, which these HS can accommodate. The net weight of ~700 grams is at least a couple hundred grams above either Intel or AMD guidelines for maximum heatsink weight, so care must be taken in transit. The bolt-through-board mounting method used for both models does go a long way to assuring high mechanical security, however.

Like the Swiftech and Alpha through-the-board mounting systems, these Thermalrights both use captive springs on machined screws to apply the correct amount of tension between the HS and CPU. The SP94 and SP97 use the same identical hardware, in fact. One set of this hardware is laid out between the two heatsinks in the photo below (no need for both sets of hardware; they are identical).


Clockwise around the heatsinks, from the to: 2 pairs of wire clips for fans, the installation guide sheet, bag of threaded inserts or anchors, washers, and spring loaded machine screws, CPU/HS support bracket for trace-size of motherboard (the anchors thread into this piece) and a tidy TIM dispenser -- enough for several applications.

The finish of the bases is excellent, but still a step below that of the Swiftechs, and half a step below the Zalmans. Fine ridges can be felt and heard with the fingernails. Fanatical readers may consider it worthwhile to lap (fine polish with sandpaper) these bases for improved smoothness, but IMO, the job of thermal interface material (TIM) is to help compensate for such imperfections. There's also evidence that a slight degree of roughness is preferable to perfect smoothness because the former provides greater contact surface area between the mating surfaces.



NOTE: The SP97 base on the left has the protective plastic sheet with which the unit is shipped. This plastic piece is secured by adhesive. Some of the adhesive remained in patches on one of the sample heatsinks after the plastic was removed. You must use pure alcohol and a cotton wad and some elbow grease to get rid of all the residue; if left, the adhesive could affect the thermal bond between HS and CPU (and probably goop it up badly).

The fins on the SP94/97 are shaped in such a way as to provide stepped notches that are perfect fits for 70mm, 80mm and 90mm fans. Two different wire clip sets are used in combination with three pairs of holes on the heatsink fins (visible in the photo above) to accommodate these difference size/thickness fans. This is the same clever design used in previous Thermalright thin-fin heatsinks.



The sturdy plain brown box favored by Thermalright haven't changed much, but it is a bit bigger for the SP94/97 than previous ones.



The packaging of all the bits and pieces is very nicely done as usual. The contents of these boxes would survive all but the worst handling in shipping transit.



The steel reinforcing plate has two sets of 4 holes so it can be used with either P4 or Socket-A. The plate has a square hard rubber pad in the center to insulate it from the motherboard PCB traces.

The Role of Heatpipes in the SP94/97

By now, just about every PC enthusiast on the planet must know at least something about heatpipes. Just to refresh your memory, here's a quick summary:

Inside a heatpipe is a liquid under low pressure (or vacuum) that boils into vapor when it absorbs heat. This vapor then condenses back into liquid at the cooler surfaces of the heatpipe and releases the heat. Heatpipes are capable of transferring a large amount of heat per a given volume of working fluid due to the phase change (liquid-to-gas-to-liquid-ad infinitum) that takes place. In layman's terms, it's vaguely like watercooling without the pump, but better. Here is a thorough, accessible explanation of heatpipes, by Thermacore.)

Normally, heatpipes are best used to move the heat away from the source to a point where it can be more easily cooled, either by forced air or by convection. A multi-finned radiator and a hot electronic component connected to each other by one or more heatpipes 6~12" long is fairly typical.

The concept in the SP94/97 is quite different: The three heatpipes are essentially being used to distribute the heat more evenly throughout the entire heatsink, especially the tips or ends of the fins. At first glance this just looks like a gimmick: Surely the thin copper fins are not long enough to benefit from the higher conduction of the heatpipes? The Heatlane Zen NCU-1000 CPU Cooler uses this type of heatpipe arrangement, but there, the distance between the base and the end of the tallest fin is 3 times that of the SP94/97. Well, the proof is in the cooling, so let's wait and see how the SP94/97 fares against the very similar SLK900, a known high performer.

Finally, there are better and worse mounting orientations for the heatpipes. A FAQ on the Thermalright website states:

Due to the multiple heatpipes we use on the design to minimize the gravitation effect on Heatpipe performance, we at Thermalright tested only 0.05 Celius/W on an open case environment for best/worst scenario with Panaflo FBA09A12H.

However, the illustration below from the instruction sheet ranks alignment from 1 to 4 stars. In any case, the best/worst difference may be larger with a low airflow fan, so adherence to this guide is strongly recommended.

TESTING

The testing for this article took place over many days. As stated earlier, the article incorporates a great many tests and measurements. Each test was run at least twice, sometimes three or four times when the results seemed unlikely. In most cases, the numbers posted here were obtained almost every time. In cases where there was more variance, it was rarely more than by 1°C, for which the numbers that best fit a trend were selected for final posting here, or an average was used.

P4 HS TEST PLATFORM

The P4 HS test platform is an open system not enclosed in a case. The test platform and procedure is described in detail in our HS Test Methodology article, but there have been some changes, which are noted below.

  • Intel P4-2.8A This is a change from the ~50W P4-1.8A used in previous P4 reviews. Given that the current fastest P4s are 3.2GHz running 800 MHz bus, it was time for a hotter and faster CPU. The "Thermal Design Power" of this P4-2.8 (533 MHz bus) is 68.4 or 69.7W depending on the version. As the CPU is a demo model without normal markings, it's not clear which version it is, so we'll round the number off to ~69W.
  • Intel D845PEBT2 motherboard - Intel 845PE Chipset; on-die CPU thermal diode monitoring
  • Panaflo FBA08A12L1A 80mm DC fan
  • Any VGA card (AGP)
  • 256 MB DDRAM
  • Any hard drive (in Smart Drive from Silicon Acoustics)
  • DigiDoc5 w/ thermal sensors
  • Any Good PSU
  • Zalman Multi-Connector (ZM-MC1) and Fanmate1 voltage controller
  • Arctic Silver Ceramique Thermal Compound
  • Two-level plywood platform with foam damping feet. Motherboard on top; most other components below. Eases heatsink changes and setup.
  • CPUBurn processor stress software
  • Intel Active Monitor and Motherboard Monitor software to show CPU temperature

Please read the caveats about P4 thermal
diode accuracy
on page 2 of the HS test methodology article. There
is evidence to indicate the diode might read >10°C lower than actual core
temperatures. At this time, we use the thermal diode to obtain CPU temperature
information -- like the vast majority of hardware sites. Other options, such
as Intel's P4 Thermal Test Vehicle (discussed in this Intel
PDF document
), are still being investigated.



Here's the P4 system set up with the threaded through-board anchors for the heatsink.



The steel reinforcing plate on the underside of the motherboard.



Image from above showing two of the spring-loaded bolts.



Reference Panaflo 80L fan placed (not clipped) on SP94, ready for testing.

Reference Panaflo Fan Summary

Table 1: Panaflo FBA08A12L1A Speed, Airflow and
Noise
VDC
12V
9V
7V
5V
RPM*
1880
1520
1230
760
dBA @ 1m**
21
17
12
10
CFM**
24
19
14
9

* RPM was measured on two Panaflo samples using MBM5 along with a custom-made tachometer board provided by Fan Control, an electronics-savvy member of the SPCR Forum. This allows accurate readings of fan speed down to much below 1000 rpm, which is usually not possible with standard on-board sensors. The readings from the two fans were averaged for the above table.

** Both [email protected] noise and CFM (cubic feet per minute) airflow figures are extrapolated from the 12V manufacturer's specifications. Please see the Excel table by gmJamez in the article Choosing Fans for Quiet, High Airflow: A Scientific Approach.

Table 2: Panaflo FBA08A12L1A Description
of Perceived Noise


Note that the description below refers to the best samples of this fan.
There are some variations from batch to batch, generally less than many
other brands and models.
12V
Plainly audible from several feet away.
Some low frequency vibration can be heard and felt; direct mounting to a
chassis accentuates the low frequency noise. Most audible aspect of sound
is centered around 5-10 KHz; whine and whoosh
are best descriptors. Much quieter and smoother sounding than most commonly
used fans.
9V
Noise is substantially reduced. Its overall
character is not changed, but the 5~6 dBA/1m decrease may be enough in many
cases to put the level below the overall noise level of the PC system (thus
making in inaudible) in a less than quiet environment (~30 dBA).
7V
All aspects of noise are greatly diminished.
From a couple feet away, it is inaudible to many people, even in a very
quiet room. Qualitatively, it is a hum rather than a whoosh
or whine. No high pitched noise at all. Enough vibration remains that direct
chassis mounting produces a bit of low frequency accentuation.
5V
Essentially inaudible in free
air except from under one foot distance. Someone with higher hearing sensitivity
might hear it from 2 feet, but it would only be a whisper, or rather a very
quiet buzz. The vibration level is low enough that direct
chassis mounting produces no significant boost in low frequency noise.

Test Procedures

  • Temperatures were taken after 20 minutes of CPUBurn.
  • The system was allowed to cool between tests for ~15 minutes. The ambient temperature was 20C throughout.
  • All temperatures in degrees Celsius.
  • Diode: reading from P4-2.8 CPU diode via Intel Hardware Monitor.
  • Temp Rise refers to the difference between ambient temperature and the diode reading. The ambient temperature during all testing: 20? Celsius., measured with an external thermometer ~6 inches above CPU HS and fan.
  • °C/W refers to the number of degrees of temperature rise per watt of heat dissipated by the CPU. As mentioned above, we'll use 69W as the CPU heat. If the power dissipation is correct, °C/W can be used to accurately predict the performance of the HS with processors of different wattage.
  • To estimate temperatures inside a PC case, add a minimum of ~10°C to the room temperature. The case temp rise can be less or more; this varies on so many factors that it is difficult to be more specific.
  • The idle temps are not reported because the highest was only 29°C.
  • The performance with the Panaflo at 5V were not reported because of the limited cooling usefulness at that voltage (all temps were above 60°C), the time required for all this testing, and the fact that in the vast majority of cases, the difference between the Panaflo at 7V and at 5V is negligible.

PERFORMANCE: SP94

Table 3: SP94 on P4-2.8
Panaflo 80L fan
Diode
Temp rise
° C/W
12V
44°C
24°C
0.35
9V
47°C
27°C
0.39
7V
55°C
35°C
0.51

Regular readers will complain that these results are not directly comparable to previous P4 heatsink tests at SPCR, and they will be right. In anticipation of such complaints, heatsinks tested in the past with the P4-1.8 were pressed into service here. These include the SLK900 and both aluminum/copper as well as all-copper versions of the Zalman 7000.

The Zalman 7000 heatsink poses a challenge to a rational comparative method. Its integrated fan differs substantially from the reference Panaflo:

Table 4: Panaflo vs Zalman 7000 fan
Model
Size
Noise
RPM
12V
5V
12V
5V
Panaflo 80L
80mm
21 dBA
9 dBA
1900
760
Zalman 7000
92mm
25 dBA
20 dBA
2400
1300

We can only guess that the airflow is at least 50% greater than that of the Panaflo. In the end, the real criteria for us is noise, so... Careful listening and confirmation with a lab-grade sound level meter showed that that noise level of the Zalman at 5V was similar in level to the Panaflo at about 9V. This was confirmed by several listeners. The noise did not sound the same, but with both, the predominant noise was that of rushing air "whooshing". In any case, our approach is that comparing the Zalman 7000 at 5V to other HS with Panaflo 80L at 9V is reasonably fair: They sound about the same.

COMPARISON: SP94 vs Others

Table 5: Just the Thermalrights
Panaflo fan
SP94
SLK900U
Temp
° C/W
Temp
° C/W
12V
44°C
0.35
44°C
0.35
9V
47°C
0.39
48°C
0.40
7V
55°C
0.51
56°C
0.52


Table 6: Just the Zalmans
Fan Voltage
Z7000alcu
Z7000cu
Temp
° C/W
Temp
° C/W
12V
36°C
0.23
37°C
0.24
9V
39°C
0.28
39°C
0.28
7V
41°C
0.30
42°C
0.32
5V
45°C
0.36
46°C
0.37

Table 7: SP94, SLK900, 7000A(alcu), 7000A(cu)
Panaflo
SP94
SLK900
Z7000alcu
Z7000cu
12V
44°C
44°C
(fan at 5V)
9V
47°C
48°C
45°C
46°C
7V
55°C
56°C
-
-

So what do tables #5, 6 and 7 above really tell us?

1) The SP94 and SLK900 appear to be very close in performance with this CPU. The SP94 seems to have a tiny edge as airflow is reduced, and although 1°C is is within the testing setup's margin of error, the difference came up the same repeatedly. Suffice it to say that for low noise cooling performance, both are excellent.

2) As in the previous full review of the Zalman HS, the all-copper version is slightly edged out in performance by the aluminum/copper hybrid. Whether this is due to the fan on the latter being slightly more efficient is hard to say, but again, 1°C is within the testing setup's margin of error.

3) The Zalman 7000s appear to perform slightly better than SP94 -- at the established noise level. This result has to be regarded with caution. The Zalman has an integrated fan that probably maximizes the available airflow, while it's possible that a different fan could produce a bit more airflow at the same noise as the Panaflo at 9V (and Zalman at 5V). If so, then the performance of the SP94 would be a higher than seen here.

4) Another consideration is that for some users, a Panaflo at 9V or the Zalman at 5V are both too loud. For users who seek a level of noise lower than that afforded by the Zalman 7000 heatsinks at 5V, the practical solutions is a heatsink without an integrated fan, like the SP94 plus a quieter fan like the Panaflo 80L at a lower voltage. Few fan controllers provide voltages below 5V.

SOCKET-A HEATSINK TESTING PLATFORM

  • AMD XP2500+ Barton core (1.83GHz, 1.65V, 68.3W, 90? C maximum core temp). This is an upgrade from the XP1600+ Palamonio core CPU used earlier. The power increase is just 5W more. According to MBM5, the motherboard delivers 1.71V for the 1.65 setting (typical of Asus, which uses this mild overvolting or overclocking for improved comparative performance) so the actual power is probably a watt or 2 higher.
  • Asus A7V8X-X VIA KT-400 chipset motherboard
  • Panaflo FBA08A12L1A 80mm - Our reference fan
  • Any VGA card (AGP)
  • 256 MB DDR SDRAM
  • Any hard drive (in Smart Drive from Silicon Acoustics) - The noise is not big a deal, but as the test platforms use noisy IBM and WD drives displaced from working PCs by Barracudas, the SD helps prevent insanity cause by whining.
  • DigiDoc5 w/ thermal sensors
  • Any Good PSU
  • Zalman Multi-Connector (ZM-MC1) and Fanmate1 voltage controller
  • Arctic Silver Ceramique Thermal Compound
  • Two-level plywood platform with foam damping feet. Motherboard on top; most other components below. Eases heatsink changes and setup.
  • CPUBurn processor stress software
  • Motherboard Monitor software to show CPU temperature
  • Plywood board with foam damping feet. All the components are mounted or place on this board for easy access and portability between two work areas. The PSU is held down with a bungie cord; the motherboard is held down with screws and insulated spacers to minimize damage during multiple HS changes.


Four holes around CPU socket populated with screw inserts which thread
into the steel reinforcing plate on the other side of the motherboard.

 



From a different angle: SP97 mounted with Panaflo fan perched on top.



SLK900 shown here mounted with through-board bolts and springs. It
is examined here on the Socket-A platform.

Mounting the Zalman 7000A on Socket-A

The photo below shows the bits used to mount the Zalman 7000 HS to a Socket-A motherboard. There are 2 aluminum brackets. The blue one goes on the side of CPU socket with the locking lever; the silver one goes on the other side. Two screws hold each bracket in place. The brown perforated thing is a small sheet of paper with cutouts that are used as grommets or gaskets!


Shown below is the Z7000CU mounted on the Socket-A motherboard/CPU. A potential weakness is that unlike the Thermalright SP94, SP97 and SLK900, the Z7000s do not use a back reinforcing plate on the other side of the motherboard. Especially for the copper version, which weighs 773 grams, this lack seems unwise. It may be prudent for 7000CU users to obtain a CPU/HS reinforcement plate (perhaps even from Thermalright?!). It would be easy to adapt a P4 reinforcement plate for this purpose. Just drill out four new holes in the correct pattern for the AMD HS mounting holes. But do be careful of shorting anything against metal on the trace side of the board.

PERFORMANCE: SP97

The ambient temperature for this test was also 20°C, and the CPU power (heat) is 68.3W.

Table 8: SP97 on XP2500+
Panaflo 80L
Diode
Temp rise
° C/W
12V
41°C
21°C
0.31
9V
46°C
26°C
0.38
7V
56°C
36°C
0.53

The results are good in comparison with results from previous HS reviews, but how does the SP97 compare with its immediate competitors?

5. COMPARISON: SP97 vs Others

Table 9: Just the Thermalrights
Panaflo fan
SP97
SLK900U
Temp
° C/W
Temp
° C/W
12V
41°C
0.31
45°C
0.37
9V
46°C
0.38
51°C
0.46
7V
56°C
0.53
58°C
0.56

The SP97 enjoys a big 4~5°C cooling performance margin with the Panaflo fan at 12~9V, and the advantage remains at 2°C down to 7V.

Table 10: Just the Zalmans
Fan Voltage
Z7000alcu
Z7000cu
Temp
° C/W
Temp
° C/W
12V
36°C
0.24
37°C
0.25
9V
39°C
0.28
39°C
0.28
7V
41°C
0.31
42°C
0.32
5V
43°C
0.34
45°C
0.37

I reiterate my earlier comment: The 7000ALCU sample probably has a slightly more efficient fan than the one on the 7000CU sample. No other explanation really makes sense. Sure, they could have the same cooling performance if the heat source does not cause thermal overload on either, but to have the hybrid outperform the all-copper? The difference must be in the fan; in the materials, the copper one has the clear theoretical superiority.

Table 11: SP97, SLK900, 7000A(alcu), 7000A(cu)
Panaflo
SP97
SLK900U
Z7000alcu
Z7000cu
12V
41°C
45°C
(fan at 5V)
9V
46°C
51°C
43°C
45°C
7V
56°C
58°C
-
-

As with the earlier comparison (tables 5, 6, 7), the Zalman 7000s were compared with the Thermalrights and Panaflo at 9V. The SP97 and Zalman 7000CU tie for second place, with the Zalman 7000 hybrid taking the lead again. The SLK900U is clearly in last place, falling behind by a big margin in the 9V category.

FINAL CONCLUSIONS

The SP94 and SP97 clearly deserve the top rank on Thermalright's product lineup. They are clearly better performers than the earlier SLK900, and the advantage will show up especially with hotter CPUs. In many ways, the performance of the Thermalright heatpipe heatsinks is limited by the exclusive use of the Panaflo 80L in this review. A quiet 92mm fan such as the Panaflo 92L may be the best compromise for these heatsinks, which are probably best used with the hottest CPUs. With the increased airflow from Panaflo 92L, I would expect both SP94 and SP97 to equal or exceed the performance of the Zalman 7000s at the latter's 5V (~20 dBA) noise level. Even with the lower airflow Panaflo 80L, the Thermalrights come very close to matching the Zalmans.

Pros

Cons

* Great basic design

* Extremely secure bolt-thorough mounting

* Very high cooling performance

* Flexible fan options, including a 92x38mm fan
* High weight

* Cost?

* Complex installation

For those who like their heatsinks and fans served a la carte, there is probably no higher performance choice than the SP94 or SP97.

* * *

Great thanks to Thermalright and Silicon Valley Compucycle (SVC) for the Thermalright heatsink samples, and to


Sharka Corp

for the Zalman samples.

* * *

An Aside: SP94's Trouble with Heatspreaders

The SP97 vs SLK900U and the SP94 vs SLK900U comparisons are really interesting. The SP97 on the XP2500+ enjoys a big 4~5°C cooling performance margin with the Panaflo fan at 12~9V, and the advantage remains at 2°C down to 7V. These numbers are quite a contrast to the SP94, which shows barely any cooling advantage over the SLK900U on the P4-2.8.

Now it's very clear that the SP94 and SP97 are essentially twins. They differ only in the interface hardware to mount them to different CPU platforms. It seems logical to think that they would have a similar advantage over the SLK900U on either platform. (Each of these four measurements were done 3 times, by the way, with the same results.)

So what is the difference? The difference appears to be in the CPU itself. Not in the power dissipation, which is purportedly within a couple percent of each other (a watt or two), but in the presence of the heat spreader in the P4-2.8 and the absence of one on the XP2500+.

It has been discussed in the pages of SilentPCReview that Intel's integrated heat spreader (IHS) has the primary functions of

  1. physically protecting the CPU core from damage while heatsinks are installed or removed, and
  2. to reduce localized hot spots on the die

Given the number of reports of cracked AMD K7 and XP core in the last few years, this was probably a very smart marketing move by Intel. AMD has finally followed suit this year, integrating a heatspreader on their 64-bit CPUs.

One side effect of the IHS is to actually increase the overall thermal resistance of the CPU. (Please see the comment entitled UPDATE on Integrated Heat Spreaders! in the article CPUs Ranked by Noise / Heat.) This effect occurs because the IHS introduces 2 additional thermal layers: The IHS itself and the interface between the inside of the IHS and the core. According to one intrepid modder who removed the IHS, a large gob of ordinary thermal interface material (TIM), probably silicone, is used under the IHS. So instead of a single TIM layer between heatsink base and CPU core, there are 2 TIM layers, with a piece of not-that-smooth or flat copper in between.

My hypothesis, then, is that the IHS on the test platform P4-2.8 increases thermal resistance enough to render meaningless the performance difference between the SP94 and the SLK900. Because the XP2500+ does not have a heat spreader, the cooling difference between the SP97 and SLK900 is clearly seen. The SP97/SLK900 comparison suggests strongly that the SP94 is similarly better than the SLK900, even though the advantage is not seen here.

There are probably two circumstances in which the cooling performance gain of the SP94 could be seen:

  1. In hotter conditions, particular with a hotter CPU.
  2. On a CPU without an IHS.

Yet Another Aside: Z7000alcu at 4V

We've deliberately limited the low voltage testing of fans to 5V, simply because it is the lowest practical voltage that's easily available in a PC. Most fan controllers also don't provide lower than 5V, partly because that's about the lowest voltage at which most 12V fan will reliably start. Zalman fan controllers all provide 5V min.

But as this review was being wrapped up, it occurred to me that I do have a simple way of trying the Zalman 7000 HSF at lower than 5V. The Sunbeam Rheobus Fan Controller reviewed here last year, actually goes from 0V to 12V output in a continuous range. It got dug out from under the clutter of the test lab and hooked up into place. The only way to know how much voltage is going to a fan is to actually measure across the terminals, which is what I did. At 4V, these are the excellent results with the Zalman 7000alcu on the XP2500+:

Voltage
CPU Temp
° C/W
4V
51°C
0.46

The noise level at 4V is much quieter than at 5V. Instead of being close to the Panaflo at 9V, it's closer to the Panaflo at 7V or less. It is tough to discern exactly, because the level IS very quiet, and to gauge differences between two quiet sources accurately, all other noise sources need to be eliminated (including fans and hard drives of the test systems, etc). This being the helter-skelter days just before the winter holidays, it is not possible to set up such testing right now, it will have to wait for another time. Nevertheless, the above result shows that cooling effectiveness of the 7000 extends down even to 4V, and this may be a viable option for 7000 users who wish the very lowest noise level.

Comment on this article in our Forums.

Sections: 

Google

www SPCR