Viewing page 4 of 5 pages. Previous 1 2 3 4 5 Next
CPU, GPU, and PSU cooling
We have now reached a crucial stage in our workstation build, namely, deciding on how to cool the three hottest components: the CPU, GPU, and PSU. It is important to choose an option that performs extremely well at full load. Workstation noise is rarely constant; it can significantly increase along with the increase in computational load. The harder the chips work, the hotter they get, and the faster the fans are required to spin. The cooling solution we choose must be able to reduce this increase in noise at full load as much as possible. In other words, the cooler must have adequate headroom, both thermally and acoustically.
If you go to a computer showroom, such as the Apple Store, you may be tempted to believe that the Mac Pro you are listening to is quiet. It might as well appear to be, but it is likely operating at a very light load and its acoustic output is being drowned out by the high level of ambient noise in the store environment. In order to truly assess the acoustics of a workstation, you'd have to buy it, install it in your lab, and run a specially designed benchmark application that stresses the CPU, GPU, and PSU at levels comparable to those you are likely to operate on in your daily research computing. I predict that you would discover that the workstation you'd just purchased gets unacceptably loud at full load. It has happened to me every single time I worked with an off-the-shelf workstation. At some point I decided that it was going to be impossible to buy a quiet workstation and that I would have to design and build one myself. I encourage you to do the same.
The stock heatsink and fan that comes with a CPU is likely to be unacceptably noisy. We have to replace it with an aftermarket heatsink and fan. The basic thermodynamics of CPU cooling are that the larger the area of the heat exchanger, the more efficiently it is going to perform. CPU heatsinks are typically made of copper or aluminium, with numerous heat pipes and fins to maximize heat exchange. They can be rather large and are likely to require additional mounting hardware. In addition to efficient cooling, large heatsinks provide an extra layer of redundancy, a crucial feature of a research workstation. Should the cooling fan fail, a large heatsink is able to provide enough passive cooling capacity to prevent the CPU (or GPU) from overheating and failing.
Figure 1. Thermalright HR-02 heatsink. Note the asymmetrical design
For the purposes of this build, I am going to use the Thermalright HR-02 heatsink (Figure 1) and a Scythe 120mm PWM fan. There are several equally good heatsinks available, but the HR-02 has the advantage of an asymmetrical design, making it more likely to fit a motherboard, with all RAM slots filled, even with RAM heatsinks installed (Figure 2). I am going to attach a 120mm Scythe fan to the heatsink with the provided pins. We want the fan to push cool air through the fins of the heatsink towards the rear case exhaust. Thus, we're going to create a sort of "push-pull" airflow configuration, which has been shown to work particularly well for heat exchange purposes.
Figure 2. Thermalright HR-02 heatsink and Scythe PWM fan. Note the "push-pull" airflow configuration and adequate clearance of the RAM modules.
In order to cool the video card, we are going to have to get rid of the stock heatsink and fan. There are a few after market solutions available. I have decided to use the Thermalright Shaman heatsink and fan for this particular build (Figure 3). The rationale is exactly the same as in the case of the CPU; we want to put a large heatsink on the chip, so it keeps the video card cool, even with a fan operating at low rotational speeds (say, up to 1,000 RPM). The Shaman is an excellent cooler, which, along with the TY-140 PWM fan, provides massive, whisper-quiet cooling to the GPU, even at full load. I must point out that both the CPU and GPU fans are pulse-width modulated and regulated by the BIOS. It is a very simple, elegant, and effective solution.
Figure 3. Thermalright Shaman VGA cooler installed onto the NVIDIA GTX 460 reference card. Note that the entire kit requires 4 PCI slots worth of space.
The power supply is typically cooled with a built-in fan. It is crucial to choose a relatively high-wattage and an 80-plus efficient power supply. The high wattage is going to provide ample, clean power to all of the components with a great deal of headroom. Such a power supply is extremely unlikely to overheat and will thus operate at a relatively low temperature. Consequently, the cooling fan is not going to have to spin very fast, and the power supply will remain quiet, even at full load. For the purposes of this build, I have chosen the Antec CP-850 power supply. It is larger than the ATX-standard PSUs, so it will fit only a handful of Antec cases. However, the benefits of high wattage, cool operation, and a low price are well worth the obvious compatibility compromise. Figure 4 shows the non-linear distribution of the fan's rotational speed, while Figure 5 illustrates PSU efficiency.
Figure 4. Duty vs. RPM curve of the Antec CP-850 power supply (courtesy of Antec)
Figure 5. The 80-plus efficiency curve of the Antec CP-850 power supply (courtesy of Antec)
The Antec P183 V3 chassis has a unique design whereby the PSU is installed in a separate chamber. It, therefore, receives additional cooling from a 120mm intake fan (Figure 6). Other cases may have a dedicated PSU fan blow hole in the bottom. You would then position the PSU with the fan facing down to allow it to intake cool air from outside the chassis. Either design should work well.
Figure 6. Antec P183 V3 chassis airflow in the bottom chamber allowing the PSU to be properly ventilated
|Help support this site, buy from one of our affiliate retailers!|