Liquid cooling expands application spectrum for CompactPCI computers
In early December 2017, I had the honor of delivering one of the keynote speeches at the DESY MicroTCA Workshop in Hamburg, Germany. The workshop was well-attended, as 183 participants from 25 institutes and 23 companies gathered to share the latest developments in MicroTCA hardware, software, and applications. The pre-workshops, held December 4 and 5, included a tutorial for MTCA.4 beginners, interoperability testing opportunities, and an introduction to ChimeraTK, a popular software suite for managing MicroTCA (and other) components. The industrial exhibition was also well attended, with 14 companies present.
A typical industrial computer based on CompactPCI Serial with a 3U European card format generates a power loss of 50 to 80 W in the form of waste heat when used with a XEON 3 processor (Figure 1). Such systems typically use high-performance graphics cards based on the NVIDIA GeForce GTX or the Quadro family for the visualization of raw data, pre-processing, and events. Again, this adds up to 150 W of power lost. It is easy to network these systems via the built-in Ethernet interfaces or the CompactPCI Serial backplane , thus further increasing the computing power. In fact, systems with 5 or 8 CPU modules are easily possible without problems. Easily? Yes, if only the high power loss of more than one kilowatt something barely bigger than a shoebox would not overwhelm normal cooling systems.
Cooling the space in autonomous vehicles
One application for such computers can now be found under the very topical heading of “autonomous driving.” All manufacturers in the automotive industry are doing research on this subject and testing vehicles out on the road. The industrial computers contained in these systems capture relevant data from which they derive the insight as to which data and algorithms are required. The vehicles under test are already able to react with these data autonomously in some instances (steering, braking, accelerating), albeit under the strict supervision of the test engineers and drivers. These test systems are typically protected in the trunk of the vehicle, along with all other system and sensor electronics. This remote, closed-in location adds a further problem to the conventional exhaust air-cooling of multiprocessor systems with high-performance fans.
The problem: How to remove the waste heat from the subframe, along with the question of what to do with the heat inside the luggage compartment, in the immediate vicinity of the other devices. Inside the trunk, mounting temperatures would soon lead to massive problems in all test systems. An accurate impression of how full it can be in such a car trunk can be gained by viewing the pictures in the article “Piloted driving with the Audi A7 Jack” on the mobilegeeks.de website .
Heat-dissipation problems can be addressed by the CoolConduct system rack from EKF, which was developed in tandem with a German specialist of industrial liquid cooling systems . Several components work in concert: A 19-inch compatible conductive plate is integrated above the subassembly of the subrack. A coolant is passed through the plate and is the primary heat exchanger for the components to be cooled (Figure 2). A unit with (redundant) cooling pumps and a coolant tank are connected to the subrack by two bayonet couplings, which can be placed several meters away from the subrack. The bayonet couplings are blow-back proof, which means that they can be connected and decoupled in the field without coolant loss.
The secondary cooling circuit is connected to the cooling pump with a further heat exchanger, which cools the coolant in the outside area. By means of these two circuits, the system can be used and cooled independently of the ambient air. Above all, the system allows no entry of dirt into the subassemblies via cooling.
Dissipating heat in a confined space
Back to our presented trunk workspace for our multiprocessor system: The primary heat exchanger cools the processor and graphics modules in the rack and transfers the heat via the medium to the secondary heat exchanger. This is mounted underneath the vehicle, transversely to the direction of travel, and transfers the heat to the environment, supported by the airflow. The entire system monitors itself, with any errors reported to the test system itself and processed.
The system uses a novel way to connect the modules to the primary heat exchanger in the subrack. The current technology uses so-called wedge locks, used primarily in the military sector, which require specially designed (read: expensive) and constantly updated assemblies for this purpose.
The CoolConduct system uses current commercial off-the-shelf (COTS) computer assemblies from the company-specific assembly program. The inverse, three-dimensional image of the assembly side is milled into a metal block and thus forms a form-tight and – most importantly – heat-tight connection specific to this assembly to the primary heat exchanger. These conductors are made of pure copper for optimum energy transport (Figure 3).
At the upper end (Figure 4) of each of these copper conductors is a planar surface, which is coupled to a special transfer module placed between the respective upper guide rails. These transfer modules are connected to the primary heat exchanger; they have high thermal efficiency and thus ensure very efficient heat transfer (Figure 5 and Figure 6).
As a result, the modules can be plugged into the subrack at any time in the field and can be easily removed for maintenance purposes, without the need for special tools and without having to disassemble the subrack.
Replacing modules on the fly
This configuration preserves one of the essential advantages of the 19-inch technology, namely the easy module replacement in the field. Often, development teams of experimental software rely precisely on this possibility to change the software or the sensor system in the field, which allows them to then continue with the experiment. Further applications of this setup are the expansion of CPU modules or the change of mass storage (Figure 7).
It is not only the high packing density and the small installation space of an industrial computer that can require special management of the waste heat. Many applications exist which require the cooling of computer systems separate from the ambient air, in order to prevent the penetration of factors like explosive gases, dust, or oil-laden air. Such harsh use conditions can be found on drilling platforms (Figure 8) as well as in open-cast mining or in the papermaking industry, where sulfuric acid air pollutants  will often result in the creeping death of modern computers, since their solder joints contain an increased amount of silver, legally required by the changeover to lead-free solder (European Union Restriction of Hazardous Substances regulation).
Additional examples of high-performance computer clusters in confined spaces are armored, self-propelled vehicles in military conflict areas; simultaneous license plate tracking across multiple highway lanes, as in homeland security applications; or facial-recognition detection in hot, dusty, or otherwise extreme environments. Such computer clusters may also be used in automated, computer-assisted waste or recycling sorting.