CPU Cooling.

Even closer still! With CPU cooling, the coolant water is brought right up to where the heat source is. The dissipated heat of the processors flows through water heat sinks with flowing water directly into the water circulation. The even distribution of the cooling water on all installed servers is ensured by two vertical distributor pipes, through which water flows after the Tichelmann circuit. Regardless of the installation height of a server, the flow resistance and therefore the flow volumes are always kept the same. One server per height unit can be connected to the water distributor; the connection is made via highgrade, very flexible plastic pipes. If these (preferably together with the wiring) are routed accordingly, a water- cooled server can even be pulled out from the server rack while operating. As the plastic pipes on the vertical distributors can be quickly and safely released, installation and removal of servers is just as problemfree as with just air cooling. The integration in the building-own cold water system is made from the raised floor or from the ceiling directly onto the two vertical distributor pipes in the rack. In contrast to the water-cooled server rack, cold water does not have to be available for the cooling. As enclosure temperatures

on the CPUs are permitted at almost 70°C, it is sufficient if cooling water at a temperature of about 50°C is available. This temperature is low enough, even in high summer, to be able to cool back with environmental air. A chiller is therefore not required in the cooling water system! This leads to enormous savings in the building infrastructure, both with the investment and with the operating costs. The option also exists of actively using the accumulating dissipated heat, in, for example, building heating via panel heating or component activation. This opens up further very significant savings potential. However, with this system not all heat sources on the server board are cooled In addition to the CPUs, there are many other smaller hotspots located here (e.g. power supplies, drives, memory cards, graphic cards, etc.), which cannot all be connected to the water cooling. Some 50% of the dissipated heat must continue to be removed from the servers and racks in the conventional method, which is logically via the conventional air cooling method. The heat load for the cooling system lies here at approximately half compared with purely air cooling; the cooling system and the RLT equipment in the computer centre can therefore be implemented accordingly smaller.

And here also, the performance restrictions of a conventional thermal management system are quickly reached. If more than 6 kW heat is released in a server, that is then, more than 3 kW of this must be diverted with air, the installation capacity of the server per rack must be reduced with relation to the geometric possibilities. Assistance is then also possible here again as a result of the additional water cooling of the rack, that is then, the combination of the two concepts described. A TCO study shows that this combination may even represent the ideal solution with regard to costrelated issues. The primary objective of this solution is to also make it possible to cool high-performance CPUs in the smallest installation space. The direct water cooling of processors also makes it possible to safely deal with 150 W and more per CPU in Pizzabox or Blade servers. Tests of up to 200 W on 10 cm2 surface area have already proven the capacity of this technology. Ultimately it is all about achieving the highest possible density of the computer power for the total building surface area. Another until now scarcely discussed potential of this solution is the option of cooling processors to very low temperatures and therefore of being able to operate drives at higher clock frequencies. Special precautions must be taken for this in order to pre vent a fall below the dew point and condensation build-up on the boards. Water-cooled servers can only be developed and put into operation in cooperation with a server manufacturer. The necessary modifications are actually quite low, however the high

requirements for safety and reliability rule out later addition solutions. As a standardised interface between water- cooled servers and racks still does not exist, the full compatibility has so far only been guaranteed among the products of one manufacturer. If no cooling water is available in a building, water-cooled servers can also be operated locally on a 19” cool back unit. Next to it in a rack for CPU cooling, any other kinds of conventionally cooled components can be operated. There are no restrictions with regard to cabling, installation and removal of components. The reliable and leakage proof version of the liquid system in the servers and racks merits special interest. The huge number of servers and processors brings many more components with just as many connection points into play. Managing everything is only possible with the exclusive use of robust, tested and proven components made in process-safe industrial production. All requirements for redundancy and monitoring will then be appropriately meet. Another essential safety feature is the HTU, a component that when built into the rack or raised floor takes over the overall cooling water supply and monitoring of a rack suite. In addition to a redundant pump group, various valves, sensors and an electronic control unit, it also contains a controlled interface to the building system. This optionally consists of a well-dimensioned heat exchanger or two magnet valves, to enable independence from the mains should there be malfunction of any kind.