Corsair H70 High-Performance Hydro CPU Cooler Review

Author:Marc Adams

Editor:Kyle Bennett

Date: Thursday , September 09, 2010

Corsair's H50 sealed processor water cooling system has been a favorite among many. While it is a very solid product there was nothing truly stellar about the H50. Corsair is back to the design table today with its new H70 sealed processor cooling system. How does it perform and is it worth the upgrade? Should be at least 20 better.


For PC enthusiasts, Corsair has become a household name synonymous with memory, power supplies, and quality. As a successful PC component company it would be easy for Corsair to rest on its laurels and stick to what it does best, but it seems that good enough is never good enough for Corsair as it continues to diversify. This is not Corsair’s first attempt at a water cooling system. That honor belongs to the Nautilus 500 which we had the pleasure of using back in the day. In fact, those units saw a lot of usage here at HardOCP as they were simple to use and easy to set up on makeshift test system. The Nautilus 500 was an external unit that worked well enough but didn't perform as well as many enthusiasts expected. Not to be deterred, Corsair is back in the ring with not one but two new complete water cooling loops, the previously released H50 and the brand new H70.

Today’s article will focus on the heavier duty H70 unit but we will compare its performance against the better known H50. We know that many of you are wondering if the H70 is on your current upgrade path. We will look at the H70’s overall performance versus other units in a following article.

If you are looking for some more coverage on the Corsair H70, you can see Kyle's hands-on experience with the H70 from August.

System Setup

Today's testing will occur a little bit differently than usual. The test bed still consists of the GIGABYTE X58-Extreme motherboard, six gigabytes of Corsair DDR3 RAM and the Intel Core i7 920. But today we are putting the system inside the NZXT Tempest EVO case. With these types of closed loop internal water cooling systems, bench testing just does not do the data or our readers justice.

Test Methods


In keeping with the spirit of the [H] we are once again doing hardware testing of all heat sinks. This means milling a very small path into an expensive CPU to place our thermocouple into. This is by far the best way to test coolers and the only way here at the HardOCP.

Temperatures for the CPU will continue to be measured using our Sperry Digital 4 Point thermometer.


For this article the GPU will be kept at stock speed to keep any excess heat away from the CPU that could impact the results. In 2D mode the 9500 GT generates very little heat and to further isolate it from the rest of the system we will install it in the secondary PCIE slot.

Thermal Paste

Noctua's NT-H1 thermal paste was selected as the paste of choice for a few key reasons. The thermal paste has been shown to provide excellent thermal conductivity allowing the heat sinks to better do their job. There is no observed curing time. That is, performance does not get any better over time. Any curing time could have introduced variables into the equation causing at best dubious results and at worst unreliable ones. Our channel milled CPU also requires a compound that is more viscous so the mating compound will not seep into the channel and run off.


Ambient temperature will be kept at 25C for the duration of the tests and measured with a MicroTemp EXP non-contact infrared thermometer and cross referenced with the Sperry Digital 4 Point thermometer. Any variance greater then 0.2C will halt the testing until temperatures return within spec for fifteen minutes.

Since we are dealing with water cooling we will allow extra time for each test to give the water in the loop enough time to reach equilibrium.


Idle temperatures will be recorded after a twenty-five minute period of inactivity. Any fluctuation during the last sixty seconds will reset the timer for an additional five minutes.


Load temperatures will be recorded after a twenty-five minute period of 100% load. To obtain this load we will be using Prime95 v25.3 set to blend mode. In this way we can heat up the CPU as well as the memory controller which is now integrated into the die. Any fluctuation during the last sixty seconds will reset the timer for an additional five minutes.


Sound levels will be measured with a Reliability Direct AR824 sound meter from a distance of four feet away. With everything turned off and the room completely silent the meter registered a sound level of 38dB(A). This is a very quiet room where a simple pin drop could be heard. All sound measurements are recorded in the very late evening to further reduce any ambient noise.