Greetings to all, and thanks in advance for each attempt at helpful answers. Before I get to my present questions, I feel it necessary to indicate both my presumptions and my problem:
The members of this forum are generally of above-average intelligence and collectively represent a diverse body of professions and expertise; however, more germane to the issue I will address is that each such member has to some degree or another had to find a way to overcome thermal issues affecting the stability and operability of their respective computational platforms.Presumption 2:
In a traditional hybrid liquid-air cooling system (in which liquid removes the heat from the device being cooled and transports that energy to a radiator which rejects that energy into air), cooling efficacy is not exactly correlated to any single factor (the most commonly cited examples probably being radiator size, water flow or air flow). Similarly, better materials do not always overcome the deficiencies of bad design (i.e.: form).Presumption 3:
Efficacy is independent from efficiency; moreover, while efficiency itself is an economic concept, it reflects the balance of competing objectives.
For instance: from the end-user's perspective, the most energy-efficient way to power a computer might be to rely on solar power, but the capital and reliability costs of such application might be prohibitively high. If operation in a developed area is assumed, grid power may be sufficiently reliable; however, if grid power is either unreliable or unavailable, an alternative energy source must be found: a petrol-powered standby generator with automatic transfer switch and a (separate) power conditioner setup might be the most efficient solution.
I am tasked with the cost-effective deployment of an austere-environment computer providing workstation performance and server reliability. As conceived, it will have limited portability. It is conceded that any solution to the problem presented will consume much more energy than the amount devoted to computing; that the ultimate solution will be far more bulky, complex and costly than a typical gaming rig; that this is not in any sense a mainstream application of technology, etc.
Here, austere-environment is taken to indicate the ambient air temperature will occasionally exceed 40C (104F) for extended periods (i.e.: days or weeks), with annual relative humidity averaging about 70 percent and ordinarily exceeding 90 percent (but occasionally being below 15 percent) during especially morning hours and ordinarily falling to 60-65 percent in the evening.
It is presently believed that remote-radiator cooling system inlet air temperature can in practice be maintained at (or perhaps a fraction of a degree below) 35C (95F); it is simultaneously acknowledged that the 21C (70F) cooling-system inlet-air temperatures common to PC-system reviews will during most of the year be impossible to match. On the other hand, there will be periods when the environmental ambient temperature is well below freezing, in which cases it is believed that recycling the waste energy will maintain system components significantly above 10C (50F).
A number of possibilities regarding how to execute such a design spring immediately to mind; perhaps the simplest in terms of execution is the air-conditioned cabinet. AFAIK, well-designed household air-conditioning ("AC") systems are able to reduce indoor temperatures by 11C (20F) or more; obviously, industrial refrigeration units are capable of much greater thermal rejection. Technically, I have the ability to assemble a R134a (or similar refrigerant)-based cooling system from separated components (dissimilar scraps, for instance), but at present I don't think such fabrication is the most efficient route to the identified goal.
Fabricating an insulated cabinet to house the principal thermal components (i.e.: the computer, less the human-interface pieces) and a household (i.e.: "window-unit") AC unit isn't a problem for me, and it seems such a solution would simultaneously solve both thermal and humidity concerns; if I need to build a controlled-humidity micro-room (e.g.: the size of a small closet) for maintenance purposes, I can. For such a system, the "ambient" air perceived by the principal thermal components would be recirculated; the airflow across the condenser would be sourced from the coolest local source and exhausted remotely.
This is only slightly different in the specifics of its implementation from the production-floor "refrigerated" systems common in manufacturing centers. It adds the advantage of very clean (recirculated filtered) air. It is assumed that I will install the electronic components under conditions of relatively low humidity, and that the evaporator will be drained; I have several ideas on how to ensure no errant moisture gets to the sensitive components.
In the alternative, piecing together a remote-radiator water-cooled system in which the radiators are mounted in a duct through which air is drawn through a filter by a centrifugal fan (aka: "blower") and exhausted remotely could lower the temperatures of most critical components to tolerable levels; filtered forced fresh air would also be used for direct application to the principal components.
The non-refrigerated-air solution would occasionally leave some components outside their certified or warranted operating-temperature range, which could lead to data corruption or other problems. However, it is common for published limits to be remarkably conservative, so I'm not certain this method is a non-starter.
Computational Platform Specifics:
2P motherboard; at present, I am leaning towards an initial configuration consisting of either a Tyan S8225 WAGM4NRF with (2x) AMD Opteron 4238 CPUs, or an Asus KGPE-D16 or a Supermicro H8DG6-F with either (2x) AMD Opteron 6212 or (1x) AMD Opteron 6220 CPUs. I am still researching RAM: the QVLs are next to useless, but as a general guide, I'm most likely going to begin with 32GB DDR3-1333 (PC3-10800). These choices should allow me to use a Radeon HD7970, which should provide good performance while being a lot less expensive than a FirePro V9800 and only a fraction the cost of a Quadro 6000.
At least for now, I will be stuck with MS Windows 7 (64-bit) for my OS.
1. Remembering that the installation environment will be as described, what (if any) cooling options have I overlooked?
2. How can I estimate the Btu load of a fully-stressed system, so as to determine what size air conditioner I ought to get for it? I realize that 1 W-hour ~ 3.413 Btu, but do I need to consider only the energy being rejected by the system in the form of heat, or should I use the estimated peak line load (ignoring the burden of the AC "chiller unit") or something else?
3. Do I need to alternately de-rate or up-rate the Btu capacity of the AC unit above or below a certain temperature and relative humidity (because that will influence the efficiency at which the condenser operates)? If so, shouldn't this information be mapped, or is reliance on a thermostat or thermometric probe sufficient? Or do I need to do the math and make a table of standardized operating practices?
4. How can I predict the optimal CFM flow rate for my system, so as to maximize the efficiency of "thermal scrubbing" both across the main board (to optimize "chill-factor" cooling arising from airflow across components) and across the evaporator coil?