Water Cooling Kit Group-Test
Published: 8th July 2013 | Source: Special Tech | Price: £142 - £175 |
Testing and Performance.
Intel i7 3960X [email protected] 1.1v (undervolted) 4.0GHz @ 1.25v 4.4GHz @ 1.35v 4.6GHz @ 1.45v Gigabyte X79 UD3 Corsair Vengeance LP Memory Corsair HX850 V2 Corsair Force GT 60GB Coolermaster Storm Trooper.
Continuity is very important in testing, and for this reason we keep as many of the potential variables as locked down as possible. We will be using OCCT in Linpack X64, AVX compatible with all logical cores tested and 90% free memory utilised. The test is set up to run automatically with just a few clicks to set it going. A 10 minute idle followed by 30 minutes of testing and a 5 minute cool down is the order of the day and brings the total test time per clock speed to 45 minutes. So as to remove subjectivity in determining whether a CPU has failed, OCCT is set to stop the test and register a fail should the max temp exceed 80 degrees. In testing we noted that if even just one of the cores exceeds 82 degrees OCCT halts the test and a fail is recorded.
All tests are conducted with the pump at the full 12 volts direct from the PSU. We have also tested the fans at the full 12 Volts and at lower speeds enabled when fed with just 7 Volts. At 45 minutes per test and 8 tests per kit that makes for a total of 6 hours per kit and a grand total of 24 hours total testing. Not content with that we also decided to test all of the systems at their highest overclock using a set of Nuctua fans. Add in assembly/disassembly and leak testing and you can see what a mammoth operation this is.
As usual we'll be testing our coolers at varying levels of overclock and increasing levels of voltage. this in turn of course means increasing levels of heat which the coolers need to dissipate. To begin with we start with the undervolted stock speed. Why undervolted? well if you have things set on "Auto", you may well be using more volts than are actually required to run at the chosen frequency, for example our 3960s will run quite happily at just 1.1volts, solid as a rock, 24/7, and as such we use this as our starting point.
It's no surprise that all the kits on test here today passed with flying colours, and with only a degree or two separating them across the board even when comparing high and low settings we get a bit of a hint that they have barely begun to stretch their legs.
Turning now to the 4GHz test we up the voltage to 1.25 volts, this is what is deemed normally as stock volts. Something we are always harping on about on the forums is AUTO does not mean stock volts, and normally if you overclocking with "auto" volts the motherboard will be upping the volts much more than needed if you were to do it manually. By whichever means it happens, upping the volts (especially from our 1.1v undervolt) does have a big impact on temps, with an average increase of 10-15 degrees seen in the results.
The 4GHz test sees the differences between the 7 volt and 12 volts temps star to widen across the test group, but again the comparative figures from one manufactures temps to another remains slight.
Upping the volts still further we achieve a stable 4.4GHz overclock at 1.35 Volts. It's here we start to separate the wheat from the chaff, with lesser coolers not able to disperse the increased heat effectively. Again we see a jump of 10 degrees or so from the figures at 4GHz. Both the H100 and the well-respected D14 are creeping into the 70s here, indicating that only the cream of the crop will excel at this level.
All the coolers on test are now into the low 40s at 12 volts and mid 40's at 7 volts, but again there is little to choose between them with only a degree separating the four at 12 volts and 2.5 degrees at 7 volts.
Finally our 4.6GHz test. Don't be fooled, this is an extreme test and the graph reflects this, you will only see the very best featured in this graph.
If we really want to measure outright performance, this is where we do it, and a quick glance will tell you that at 12 volts the ALphaCool NexXxoS wins by a few fractions of a degree from the EKWB and Phobya systems, with the XSPC kit some way behind. Dropping the fans down to 7 volts and this time the thicker radiator of the Phobya kit causes it to take the Laurels with the other three coming close together a few degrees behind.
Testing With Noctua NF-F12 fans
As both the fan speed and radiator characteristics can greatly affect the cooling performance of a given system, we thought we'd undertake a further test to help better define the radiator performance of each kit. We've done this by swapping out the kit fans for a set of Noctua NF-F12s, which are, it has to be said, one of the best regarded fans on the market.
So did swapping the fans change the results? Across the board temps were 3-4 degrees lower with the Noctuas than with the kit fans at 12 volts, with the variance being even less when things were stepped down to 7 volts. The Phobya though actually performed better with its own fans at 7 volts and about the same at 12 volts. The winner here though was EKWB, taking the crown from AlphaCool at 12 volts and pretty much sharing it at 7 volts. What is surprising is that the Phobya, with a rad almost twice as thick as the EK couldn't better it. Just goes to show that bigger isn't always better.