Introduction:

Thermalright is well-known to overclockers for their extremely popular thin-fin copper heatsinks.  The combination of Thermalright SK-6 and Delta Black Label fan kept many socket A CPUs cool a year or two ago, and Thermalright has continuously released new heatsinks to improve performance.  KnightElite recently purchased one of Thermalright's newest heatsinks, the 947U, and kindly loaned it to me for  testing.

As you can see from the picture, the 947U comes with a variety of pieces and parts.  It has the typical Thermalright clips for holding fans in place, in 70mm, 80mm, and 92mm sizes. One difference in this heatsink is the use of the 4 motherboard mounting holes for mounting to Socket A or Socket 478 systems.  This is always welcome to me.  Thermalright also includes clips if you want to use the stock Intel P4 mounting system. The mounting system is somewhat unique: A Baseplate mounts on the back of the motherboard and standoffs are screwed into this Baseplate. The heatsink then uses springs on bolts to ensure proper mounting pressure.

It is pretty clear when one looks at the 947U that it is similar in design to the SK6 and all of the other thin fin copper Thermalright heatsinks. You'd also guess that all that copper is pretty heavy, and you'd be correct (570g).  That isn't even close to AMD specs, but since it uses the motherboard holes for mounting I am not too concerned it's going anywhere.  It's also a pretty large unit and may not fit on all motherboards.  You should check for compatibility before making an impulse buy. In the picture you can see a variety of holes in the fins for the various fan clips. Their placement allows you to use 35 or 25mm thick 80 and 92mm fans on the 947U. I didn't test any 70mm fans but presumably one of these holes works with them as well).

The build quality and finish of the 947U is quite good throughout, and the base appeared to be fairly well done.  There are faint grooves visible on the surface and so an expert lapper might improve the finish slightly (if they could keep the base flat while lapping).  You can see from the picture that the baseplate is quite narrow; felt pads are included to ensure that the heatsink sets flat on AMD CPUs.  It's too bad every manufacturer hasn't taken Swiftech's lead in baseplate finishing; it really does make a difference.

Installation and Ease of Use:

I make it no secret that I prefer using the AMD mounting holes.  In all fairness, though, most 4 hole mounting solutions still leave a little to be desired. The standoffs have a tendency to come loose when remounting the heatsink and falling back behind the motherboard. It's a pain to have to remove motherboards to find the nuts and retighten them.  The mounting bracket that Thermalright uses solved this issue for me.  I used a nut driver that I had in my tool chest that had a hex socket the same size as the standoff, and I was able to tighten the standoff down well enough that it did not come loose after 7 heatsink remounts.  If I were building something similar though I'd use lock nuts to ensure that the standoffs NEVER come loose by accident.  My only annoyance with this heatsink involved the washers that go on the end of the springs for bolting the heatsink down. These little thin metal washers are not a great idea to use in this application as if they fall off the end of the screws and are lost (believe me it can happen) then they could short something out. I suggest that you superglue these in place on the black hold down to save you the headache of looking for them in your system.

Testing Methodology:

A variety of tests were performed to gauge the heatsink's performance on an AMD system. An Epox 8K3A motherboard was used as the test platform after modification of the socket pins to remove the contacts of the motherboard's internal diode reading circuitry. CPU diode readings were obtained by soldering 5" of twisted pair copper wire to (a) the diode pins of a 1700+ JIUHB stepping Thoroughbred B processor and (b) a Maxim MAX6655 evaluation system (0.125C resolution). The air temperature 1" from the intake of the heatsink fans was measured at a slight offset from the center of the fan with a Digitec 5810 dual linear thermistor (0.01C res) with a YSI 703 probe.

A special formulation of thermal paste (termed "BillA Goop") was used for all tests. This paste is designed to minimize the settling time required for consistent temperature measurements. Heatsink mounting was repeated five times, as follows. Heatsink base and CPU were cleaned with a lint-free cloth to remove paste, paste was reapplied, heatsink was remounted, and the PC was allowed to idle at 2200MHz, 1.85V for exactly 12 hours prior to testing. The processor was then placed under load using K7Burn in High Priority. Both CPU and intake air temperatures were collected after 30 minutes of load.  CPU power under load was estimated at 81.6W for this MHz and Vcore from tests where deltaT across a waterblock at different flow rates were measured at steady-state. As I have estimated "W" for the processor used in testing, I can now calculate a C/W value equal to (CPU temperature - intake air temperature) / CPU power (W).

To test the performance of the 947U over a wide range of CFM, the 92mm Vantec Tornado fan was controlled via PWM to run at 5, 10, 25, 35, 50, 75, and 100% duty using a Crystalfontz 633. 30 minutes were taken for each fan speed. The performance of several other fans was also tested with the Thermalright 947U: 80mm Panaflo L1A, NMB, Vantec Tornado, and 92mm Vantec Stealth, Panaflo M1A, and Papst . Specifications of all the fans tested are included in a table with the test results. Fans were mounted by swapping out the clips with care taken to not affect the TIM joint.

Test Results:

One of the most useful tests to perform with a heatsink is to learn how it performs with a variety of common fans. The goal is to give readers the ability to choose a target noise level or performance level and purchase fans accordingly. The results of the 947U's performance with fans ranging from 24 to 119 CFM are shown in the table below.

There are two items of note on this table. First of all, unlike the Swiftech MCX462-V, the 947U DOES benefit from the 92mm Tornado fan. There are a couple of reasons that this could be the case: heat may be more effectively spread to outer fins in the Thermalright than it is in the Swiftech, the open base of the Thermalright may make secondary cooling of processor top and motherboard an important cooling mechanism, or it may be that the way fans are clipped onto the top of the Thermalright makes for a situation where it is relatively difficult for air to be forced through the fins.  If so then perhaps higher static pressure from the 92mm Tornado can improve performance regardless of its larger dead zone. In any case, if you want the absolute best performance at any noise level then the 92mm Vantec Tornado is the fan for you.

As in the MCX462-V tests, I varied PWM% while testing the 947U with the 92mm Tornado.  I then estimated what %PWM would yield the same performance as was observed for all the other fans tested. Then I was able to plot the performance of the various fans on the 92mm Tornado PWM% graph.

When this is done then it becomes clear where the various fans stack up compared to more powerful units. The 80mm NMB fan fares better than the 92mm Stealth, but both are really too weak to get maximum performance out of the 947U. The 92mm Panaflo and Papst do much better, and the Panaflo M1A is a pretty good mate for the 947U due to its relatively low noise (30 dBA) and good performance.

I decided to investigate the performance difference between the 80mm and 92mm Tornado fans in more detail by also testing the 80mm Tornado as a function of PWM%. It turns out that the 92mm Tornado is only superior at higher PWM levels; around 35% PWM the 80mm and 92mm Tornado fans perform the same and at lower than 35% the 80mm fan provides better performance.

Finally, I want to graph the performance of the 947U vs. that of the Swiftech MCX462-V so that you can clearly relate their performance to one another. When this is done you can clearly see that the Thermalright 947U is a top performer.  The MCX462-V is one of the better heatsinks I have used, but the 947U clearly does a better job over all tested fan PWM%. Rather surprisingly, it increases its lead at low fan speeds.  My instinct would have been that the Swiftech would perform better with low airflow.

Conclusions:

The combination of an extremely well-done mounting system, high quality construction, and excellent performance make the Thermalright 947U a good choice for AMD cooling.  Thermalright managed to really put together a winner in my opinion and I can recommend it to those of you looking for an AMD heatsink that mounts through the motherboard's socket holes.  Of the fans tested, I would recommend the 80mm NMB for quiet computing duty; the 92mm Stealth simply isn't up to the task of pushing air through the 947U.  I would suggest that typical users should pair this heatsink with either a 45-60CFM 92mm or a 35-40CFM 80mm fan for best mix of noise/performance. For extreme (read deaf) overclockers then the combination of a Vantec Tornado 92mm will give you best results.

If you have any comments or Questions please email me at pHaestus@ProCooling.com