Component based liquid cooling

Component based liquid cooling

Hello all.

 

First of all sorry for posting here, but it is due to a lack of other forums which cover the area of interest related to my questions.

 

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.

 

Sincerely,

Rolf

10 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

Hi Rolf,

You may want to  post your inquiry here: https://communities.intel.com/community/tech/servers

Regards, Hal G
Intel(R) Developer Zone Support

http://software.intel.com
*Other names and brands may be claimed as the property of others.

 

 

 

>>>s 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?>>>

What do you mean by writing internal temperature? Are you referring to the server components?

Cita:

iliyapolak escribió:

What do you mean by writing internal temperature? Are you referring to the server components?

 

Yes, and the internal air temperature. Obviously I would like to know if I correctly assume that the temperature of the un-cooled components and the airtemperature inside the server housing would likely reach a workable constant temperature.

Check  http://www.frozencpu.com/cat/l1/g30/Liquid_Cooling.html    for available, run-of-the-mill liquid cooled PC. There are many other similar products.

Cita:

Rolf B. escribió:

Quote:

iliyapolak wrote:

What do you mean by writing internal temperature? Are you referring to the server components?

 

 

Yes, and the internal air temperature. Obviously I would like to know if I correctly assume that the temperature of the un-cooled components and the airtemperature inside the server housing would likely reach a workable constant temperature.

I suppose that your question is related mostly to thermodynamics of computer components.

Please read following article:

http://cba.mit.edu/docs/papers/96.isj.ent.pdf

Hi and thanks for your feedback. None of this answers my questions though. I already have found a lot of information on thermodynamics and lots of theory. What I am looking for however is answers related to which actual components in a server really need cooling and which components do not need it as long as the temperature remains stable. IE, if you selectively cool the CPUs and power supply, other components may still warm up, but maybe not beyond a critical value. Which components will go beyond the critical temperature? All? None? Just the memory chips or something completely different?

If I selectively cool just the CPUs and the power supply, disable ALL the fans and seal off the server entirely. No air comes in or out. Will the uncooled components keep warming up beyond an acceptable temperature or would it eventually reach a steady level, acceptable for server operation?

Thanks for your input!

Cheers!

>> disable ALL the fans and seal off the server entirely... >>  That would surely burn out memory, USB, Ethernet, etc.

Once upon a time, my company used to sell PC-based systems, installed directly on the manufacturing floor, with ambient temperatures 45-65 C.  To cool the PC, we used compressed air and an expansion valve (when the air expands, it cools). Most  factories, world-wide, have compressed-air supply.

A better solution, suitable for higher thermal loads, is a liquid-cooled PC: The internal ventilation circulates cooled air (cooled by a liquid-cooled heat exchanger), plus direct liquid-cooling of major heat sources like CPU, HD and graphic card.  Air flow suffice to cool all the other components.

Which is better - compressed air or water-cooled?  Compressed air is much simpler and safer, but the air must be dry.  A badly installed water-cooled system will burn itself out with a splendid show of lights. Define you ambient temperatures, thermal load and budget, then contact the vendors. I think 1000$ would suffice.

 

>> Which is better - compressed air or water-cooled? >> I would say neither when it comes to long term continuous operation. I do not intend to use water since it may fry the system. I am looking at an oil based cooling similar to the immersion approach, but more targeted to the individual components. The setup with water cooling is similar to what I am preparing for. The main difference is that I am working with oil (dielectric fluid), an external heat exchanger (which cools in this case with seawater) and a set of redundant circulation pumps. 

Another possibility I am looking at is to build a small oil-based heat exchanger 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. On the other hand, if you look at fanless and embedded systems, similar chips are being used without any cooling at all. 

>>>What I am looking for however is answers related to which actual components in a server really need cooling and which components do not need it as long as the temperature remains stable. IE, if you selectively cool the CPUs and power supply, other components may still warm up, but maybe not beyond a critical value. Which components will go beyond the critical temperature? All? None? Just the memory chips or something completely different?>>>

I suppose that various components will behave accordingly to heat transfer law. Here I mention hypothetical situation where the cooling is not present. Probably the governing value will be that related to ambient temperature inside the server case.

I suppose that for example on board VGA chip temperature can be directly related to the infinitesimal load on the device itself which in turn can be dependent on user activity like rendering OS GUI  (if any is present). So in such a case temperature of graphic chip will be related to the interaction between the device itself and the job scheduled by the CPU.

Login to leave a comment.