CPU water cooling

Water cooling is the best way to improve the cooling of your PC's hot-running components for better overclocking or less noise. James Gorbold guides you through the twists and turns involved in building the ultimate water-cooled PC.

The main selling point of water cooling is the ability to cool your PC so well that its components can be overclocked further than would be possible with just air cooling. For this reason, we built a new thermal test rig to evaluate all the water-cooling kits. This included a quad-core Intel Core 2 Extreme QX9770, which we overclocked from 3.2GHz to 4GHz by boosting its multiplier from 8 to 10. To ensure that the system was stable, we overvolted the CPU to a heady 1.525V. This is slightly more than this CPU needs, but we wanted to ensure that the water-cooling kits were placed under a lot of stress during testing.

The CPU was installed in an Asus P5K Pro motherboard based on the Intel P35 chipset. We chose this motherboard as it's one of the few decent overclocking motherboards that have separate Northbridge and VRM heatsinks, so we could use it to test the Northbridge waterblocks supplied with some of the kits. Several kits were also supplied with a waterblock for the Asus GeForce EN8800GT TOP 512MB graphics card installed in our test rig. Although this isn't the hottest graphics card available, there's no doubt that a 512MB GeForce 8800 GT is the card to buy right now as it offers superb value for money. Finally, as nobody likes an untidy PC, we needed a huge case in which to install all the components, so we selected the awesome SilverStone TJ07B-W, which has sufficient space for up to a quad 120mm-fan radiator without the need for modification.

To ensure that each water-cooling kit was thoroughly tested, we ran two copies of the dual-core smallfft stress test in Orthos until the CPU temperature reached a plateau, or it became too hot and shut down. While Orthos was running in the background, we also looped the Deep Freeze test from 3DMark06.

To monitor the change in CPU temperature we used CoreTemp to log the output from the DTS (Digital Thermal Sensor) embedded in the CPU core, while the GPU temperature was logged using Nvidia nTune. However, the raw information from these sensors isn't a whole lot of use, as the readings will vary if you use a different motherboard, or even a different BIOS revision. For this reason, the graphs opposite don't show the absolute temperature readings, but the temperature readings relative to our reference setup.

This reference setup was cooled using the best possible CPU HSF available, the winner of our last HSF Labs test, the Tuniq Tower 120 (see Issue 44, p82). However, even this awesome HSF struggled to cool the overclocked CPU in the thermal test rig. With the fan at maximum speed, the temperature was only 1ûC below the thermal cut-off temperature of the CPU. We certainly wouldn't recommend running a CPU at such a high temperature over an extended period, and it shows that if you want to overclock a quad-core CPU to 4GHz or higher, you need to water-cool it. In addition, the 120mm fan in the Tuniq Tower 120 was so loud at maximum speed that none of the Custom PC staff was prepared to spend more than a few minutes next to the test rig. The GPU temperatures shown on the graphs are relative to the standard Nvidia HSF fitted to the Asus GeForce EN8800GT TOP 512MB. Unfortunately, we weren't able to monitor the Northbridge temperature accurately, so we decided not to quote any readings for this component, even when it was water-cooled.

Scoring

Included in each water-cooling kit review are several separate score boxes for CPU cooling, GPU cooling (where appropriate), design, value and an overall score.

The cooling scores indicate how well each kit cooled the CPU or GPU compared with the Tuniq Tower 120 and standard Nvidia HSF fitted to the 8800 GT gaphics card, both of which scored an arbitrary 50 per cent. The design score was calculated by allocating a certain number of points for each of the key design characteristics. For example, we penalised kits that weren't supplied with clear documentation, or that had particularly noisy fans or pumps. The design score also factors in elements such as variable-speed fans and the mounting mechanisms for the individual components that comprise each kit. The value score is determined by adding the weighted cooling and design scores, and dividing by the price. The overall score is also weighted, and not an equal average of the initial scores.