Hello all. I am developing a server system for a specific industrial environment with poor air quality, fluctuating temperatures and humidity and scarce climate control. I am investigating the possibilities for building an air-tight high performance and fully redundant server system for virtualization with the focus on high availability critical (actual life and death) applications.
I am looking at the possibility for a cooling system which does not include air but liquid. Obviously this brings several challenges and I am curious about how realistic my way of thinking really is.
I am considering component based cooling which means that I intend to cool the parts on the server for which cooling is needed instead of cooling the entire server. So far I can identify the CPUs and the power supply. I am already using SSD drives to eliminate any unnecessary moving parts and heat.
As far as I know, most parts in a server would fail due to temperature fluctuations. However, when a server is up and running for a while and remains so indefinitely, is it correct to assume that the internal temperature would reach a workable constant while keeping the CPUs and power supply on a constant temperature using targeted liquid cooling?
Furthermore I would like to know if there are any other components that I should take into account which will heat up too much and thus can influence the availability of the server.
Any thoughts on this subject are very welcome.
A few folks are researching in that direction.
The bottom line is you need to get the heat out of the closed system.
Processor, Voltage regulators and IC always come ot mind as the hot spots, but every electronic component that passes current, generates some heat. That heat has to go someplace.
Here is one interesting approch. http://www.maximumpc.com/article/news/intel_experiments_oil_immersion_cooling_technology Intel Experiments with Oil Immersion Cooling Technology | Maximum PC
Hi and thanks for the reply. I am aware of these developments and it is an inspiring example indeed. This is however a very cost intensive approach and demands a lot when it comes to the server housing and physical location.
What I am looking for is a simplified and more targeted approach. Instead of immersing the entire server, I just want to look at which components need to be cooled. I intend to use similar oil and circulate it through custom made heat exchangers which fit on specific components (take off the heat sink and replace it with the oil based cooling element). This way the heat gets out of the component efficiently.
What I am curious about is whether I need interfaces for the entire server board and all components attached to it, or if I can get away with a simpler approach to only cool specific components.
Another possibility I am looking at is to build a small oil-based heat exchanger (think car radiator, but with oil) inside the server and to slowly circulate the air through this heat exchanger, just for the components which are not as intensive when it comes to heating (slowly as to optimally cool the air, but enough to keep circulation going). Of course in combination with the targeted cooling.
Hoping for some discussion on this topic...