Radiator Showdown! PA120.2 VS PA120.4

Abstract:

Suggestions from prominent liquid cooling experts indicate a larger radiator will yield greater results; somewhere on the order of 10-25% better per 120mm addition. An example is going from a 3x120mm radiator to a 4x120mm radiator will provide an decrease of temperatures by up to 25%.

To challenge this an experiment was derived to test this hypothesis. The outcome is shocking to say the least and will encourage you to make a wiser decision about purchasing the right size radiator for you.

THE CHALLENGE:

Can a Thermochill PA120.2 hold up to a PA120.4? Will the 120.4 crush the 120.2? What will be the percent difference? Both are respectable radiators amongst watercooling circles and both highly recommended, lets put them both to the test.

THE PARTS:

Component list:
i7 940 @ 4.0GHz w/1.45V
6GB Corsair Dominator GT's @ 1745MHz @ 1.65V
Evga X58(vanilla E758) motherboard
Evga GTX470(vanilla) gpu
Corsair 1KW HX power supply
Corsair Obsidian case

Liquid Cooling list:
EK Supreme HF cpu block
EK motherboard block
Swiftech MCW80 gpu block
Laing D5 pump w/EK Rev2 top
Thermochill PA120.4 w/4x Xigmatek Crystal white LED(1500rpm),2x LianLi 800 rpm, and 1 Cooler Master 2000rpm@50%.
High Quality pic of layout: http://www.techpowerup.org/uploaded.php?file=100810/BWloop.jpg

Thermochill PA120.2 w/2x Sycthe S-Flex 1900rpm, 1 Noctua NF-P14, 1 half broken S-Flex(2 blades caught my finger and broke)
High Quality pic of layout: http://www.techpowerup.org/uploaded.php?file=100810/102_2247.jpg
XSPC single 5.25" bay reservoir
Bitspower 45º 1/2" Compression fittings
Tygon 1/2"ID 3/4"OD tubing

THE METHOD/TEST:

Max temps from running FurMark and RealTemp's Sensor Test(Prime 95) were recorded during an approximate 5-10min run. Screenshots from each experiment were uploaded, but may be in link form to save some space. Ambient temps are at 75ºF or 24ºC, no additional fans were provided just standard central air conditioning. The test was done with the case shut with just the 3x140mm Obsidian fans and radiator fans accordingly(see pic for details). No additional applications were running during the tests, standard programs like TPU Capture, Real Temp, GPU-Z, FurMark, MSI Afterburner, and Evga Precision/OC Tuner were used. Latest versions of all were used during these tests.

FurMark v1.8.2 running 1680x1050 and 32xMSAA in a windowed screen just burns up any gpu. In fact this exact setup uses the maximum amount of gpu2 during these stress tests. Other variants of higher resolutions only resulted in a lower percentage use for the second gpu. Granted this only applies to the second part, but very interesting.

Real Temp is the best Intel temp monitoring program available. It's latest build has included the stress testing software Prime 95, a notorious stressing software application that runs all available threads at 100% and will stress the cpu and memory. The sensor test from RT will load at 100% and decrease by 10% over a 10 minute period; I initiated the test and ran the entire 100% of the first test(100% load) and then reset the test to simulate two 100% passes..

PART 1: CPU, board, and 1 GPU


PA120.4 RESULTS (Format: gpu/cpu)

Idle: 39/41
Individual load: 47/71
Full Load: 53/77

Not bad right? Temps are reasonable considering the cpu's voltage. Now lets over-volt the 470 and crank up the clocks. Max voltage through MSI Afterburner was 1087mV and clocks were set to 778MHz core and 952MHz memory. The overclock is impressive and was used during both tests.

Idle: 40/41
Individual load: 57/71
Full Load: 60/78

Ok so that seems about right. The gpu block is starting to show it's weakness and the NB is getting hotter as well. These temps are still well within acceptable range for the components. These results don't leave much room for the PA120.2 to perform in, afterall that overclock/over-volt is hefty and surely the power of roughly 500W cant be dissipated that well through a radiator at exactly half it's size... can it?

PA120.2 RESULTS:

Idle: 40/45
Individual load: 49/72
Full load: 53/77

What? Something's got to be wrong here? Well the fans, yeah the fans are 2x the performance of the 7 fans on the 120.4... wait that cant be. Well, this was an absolute fluke there's no way it can handle the 470 over-volt/overclock. The 120.2 will just crap out with those possibilities.

Idle: 40/45
Individual load: 55/76
Full Load: 61/79

WHAT???? WHAT?

Thats insane! Full load temps barely affected the gpu temps! A gain of 8/2º(respectively) is nuts.

UNCERTAINTY:

This experiment does have uncertainty; the specific value was not calculated to a standard 95% confidence level per each individual reading. The procedure for calculating the uncertainty for each value would require 5 measurements. The standard deviation of the results would then be multiplied by 1.39(the 95% confidence level) and thus given an uncertainty value. Basically a plus or minus value for each value. An estimated uncertainty value was 4ºC. That value was calculated from multiple idle readings of the gpu: 37,42,38,42,39; as you can see from some of the idle pictures showing small fluctuation. So at any times the actual idle gpu temps are within a range of 39 + or -4º.

THE LAZY FACTOR:

So heres the part where all my testing falls apart. I'm just a guy that did this for fun, but tried to keep the experiment as close to accurate as possible. There are some faults I'd like to point out:

1) Max temps were taken from 10 minutes of load. I didn't record an epic 8hr stress test, that is just a waste of my time and I think the test still holds validity. Intense tests like FurMark and Prime 95 will cause immediate crashes for any instability and give over-realistic temperatures during smaller durations. It's minor but noteworthy.

Update~Part 2 shows that a longer load/stress test will not significantly increase the temperature. By the end of a 10 minute stress test you will have reached within an acceptable 90% range of your MAX load temp.

2) The rad placement of the 120.2 was outside the case and used different fans. Yes, it's outside the case and does use "better" fans. But I'm not disassembling my setup just to test this and I feel the test still remains accurate to the desired results. The fan debate is petty in my opinion and just not worth it to see how each rad would perform with 100% identical placement, fans, and length of tubing used.

Update~Again part 2 shows that mounting both rads internally makes minimal difference on the outcome. Even using identical fans really makes no real difference. Granted I'm not using 3000rpm Ultra Kazes, and better fans may help indicated better performance BUT this review is focused on the radiators only. As long as I'm not using a large difference in preforming fans I'll consider this issue a wash.

3) I'm not a professional. Yup, I'm not getting paid to do this. In fact I paid full retail for these parts(radiators). Yeah I paid to do this test and I have every right to test/abuse/use them how I want. I'm not out to supply Thermochill with any sort of recommendation or review, and honestly this post will probably only be read by less than 200 people. I did it because I wanted to.

Update~I'm still not a professional, I'd be honored to review and experiment with liquid cooling parts but I'm just a guy doing this for fun.

ANALYSIS:

The results are shocking to say the least. Expectations of 10-25% improvements were barely achieved. Infact the overall DeltaT(change in temp) is actually less with the PA120.2. Meaning there is less temperature fluctuation with the smaller radiator. Also, this little PA120.2 is a BEAST! I mean for it to hold temps under 80ºC for an i7 with 1.45V is great, not to mention disbursing an additional 300+W of power from the board and 470.

There are some scenarios I didn't cover which may disclose the potential of the larger 120.4. If I had the time I would run the cpu at stock speeds and probably a touch less voltage. With an undervolted cpu the NB and Mosfets wouldn't create nearly so much heat that is passed on to the graphics card. I would venture to say that is where the larger radiator would out perform the smaller. Of course there are slight variations that could be made, all rabbits holes to chase for tiny gains.

The largest point I'd like to make is that with the setup I have the radiator isn't bossing the temps around like I first thought. I thought for sure, bigger is better; however; with the testing I've concluded that there are more important components in the loop that really dictate the temperature levels or DeltaT values.

PART 2: 2 GTX470's

Part 2 is dedicated to testing the exact same scenario but with the loop dedicated to two GTX470's. I've experimented in cooling two of the hottest video cards with both the PA120.4 and PA120.2 in both stock scenarios and overvolted/overclocked. The cards are placed in the PCIE slots as the bottom card is directly under the top card. This further constitutes more heat, at least to the top gpu.

For this test I have removed the cpu, NB, and board blocks to focus the cooling directly to the video cards. I am using my trusty Cooler Master Hyper 212+, a $30 cooler that does a great job at cooling an i7 effectively. The processor overclock has been removed and voltages have been reset to factory defaults. This does give a small advantage to the internally mounted radiators as no additional heat is being generated.

Attempting to achieve conformity and accurate results I have chosen to use the same fans on both setups. The 120.4 will remain using 4 Xigmatek Crystal 120mm fans pushing with 2 800rpm Lian Li and 1 Cooler Master 2000rpm fans pulling; all the fans were operating at 100% during testing. The 120.2 is being tested with just 2 Xigmatek Crystal's pushing and 2 800rpm Lain Li fans pulling. I believe this to effectively 'level' the playing field and reduce one variable from the equation that could be a point for error.

I have also decided to allow the burn test to run longer. Prior a test of roughly 10 minutes was used, this time 10 minutes is used as the minimum. I decided to allow 15 minutes for the PA120.2 during it's overvolted/overclocked test. A note on this should be to notice the logarithmic curve from FurMark. During the longer tests the temperature increase is minimal over the last half period of time; this means that the temperature increases are going to reach 90% of it's potentially highest peak during the first few minutes. I think this validates the first test for which shorter times were used.


THE RESULTS:

PA120.4 RESULTS

Idle: 31
Load: 49
Load OC: 56

Not bad, that range of 31-56, 25º is very acceptable. Max load temps are under 60ºC and the radiator is really distributing the heat well. I'm very pleased with this setup and results.

PA120.2 RESULTS:

Idle: 31
Load: 49
Load OC: 61

Wow, again this 120.2 really shines. It holds a DeltaT of 30º, just 5º difference to a radiator exactly twice it's size. I'm speechless at how this little radiator can just handle nearly 600W of power from two of the hottest video cards on the market with little effort(aka crappy fans). This compact radiator really outperforms my expectations and is no question one of the best radiators on the market today.





CONCLUSION:


I would like to thank the reader at this point for viewing my findings. I wouldn't be doing this just for myself.

I would also like to spark a discussion, please feel free to critique my testing/methods, choice of parts, or even layout/design. Mostly I would like to spark discussion regarding the results. Feel free to discuss accuracy, expectations, or integrity of my experiment.

Lastly I would like to thank all the liquid cooling manufactures to which I have purchased their products. I have enjoyed all the products thoroughly, thank you.

Pictures:


PART 1:
Internal PA120.4:



External PA120.2:



Note the 2xSycthe on top pulling air up and the Noctua NF-P14 sticking out oddly. The bottom Sycthe is being held in place with a single screw and a rubber band

Part 2:

Picture of the 120.2 mounted internally:

Hey guys want to wager on a second experiment I'm planning?

I'm going to setup one loop with just the PA120.2 for both my 470's. Anyone thing they can guess the MAX temp for either card?

Remember that would be equivalent to nearly 600W of power being dissipated by a 2x120mm radiator.

I want everyone to know how well watercooling can be for the GF100 and gpus in general.

Updated with results of dual gpu testing with both the 120.4 and 120.2. Results again are shocking, 600W of heat being dissipated and not even a sweat.

Guess I'll have to reload everything(cpu, board, and both gpus) in one loop and see how the 120.2 fares. It just eats up everything I have and spits it back out.