Asetek data center liquid cooling technology is the result of numerous patent protected innovations and components designed to work together. Many of the server components are based on the evolution of proven desktop PC designs combined with real world feedback from million of units installed and in use. However, the more stringent mission critical demands of the data center require that components are subject to their own set of tests to match the performance and reliability needs of the data center and server markets. Asetek does not market individual components to the server and data center markets, but rather, Asetek works directly with OEMs to define, develop and integrate the components into whole liquid cooling solutions specific to their needs.
At the heart of Asetek liquid cooling architecture is the Direct to Chip (D2C) CPU Cooler. The CPU Cooler is a patented integrated pump and cold plate assembly used to cool server CPUs. Integration of the cold plate with the pump has numerous benefits. First, integration reduces connections and improves reliability. Close coupling of the pump and cold plate enables simultaneous optimization of these two critical components to provide the best possible performance.
Integration also optimizes space within the server. The low-profile integrated pump / cold plate assembly is a direct drop-in replacement for air-cooled heat sinks having the same mounting mechanisms. This allows OEMs to exploit air cooled server designs in creating liquid cooled versions for higher wattage CPUs than can be supported with air cooling. Placing the pumps within the server enables Asetek’s highly reliable low pressure distributed pumping architecture (2.5 – 4psi typical).
Because a single pump has sufficient power to circulate cooling water in a server node, servers with more than one CPU have multiple pumps in series, offering built in redundancy. Asetek's distributed pumping strategy is also self provisioning. Each time a server is added, the corresponding pump provides the flow needed to cool that server.
Heat Distribution & Rejection
While Asetek direct-to-chip technology targets the capture of heat from the hot spots in servers, it also provides options for the transporting and management of this heat to a suitable point for rejection. Be it CPUs, GPUs or memory, suitable heat rejection points range from within a server’s normal cooling air stream using Asetek ServerLSL™ to outside of the data center building using Asetek RackCDU™. At the extreme, heat rejection translates to heat re-use with Asetek liquid cooling being used for the heating of buildings.
In-RackCDU D2C™ provides the option of having the RackCDU mounted in the server rack along with the servers themselves. This can be desirable when it is important not to take up any additional aisle space. Mounted in the top 3U or 4U of the rack, InRackCDU functions in the same manner as the VerticalRackCDU.
Vertical-RackCDU™ is a (Cooling Distribution Unit) mounted in a 10.5-inch rack extension to the rear of a server rack. Heat exchangers in the RackCDU remove the heat captured by the server coolers transferring it to facilities water. Heat is exchanged with no mixing of facilities and server cooling water. The final heat rejection point can be dry coolers, cooling towers or other such standard HVAC units.
GPU Accelerator (GPGPU) / Coprocessor Cooling
Asetek direct-to-chip liquid cooling focuses on removing heat from the hottest locations in servers. GPUs and other coprocessors such as GPGPU (General Purpose Graphic Processor Units) are a growing hot spot in high performance servers as manufactures offload processor intensive tasks from the main processor for more performance.
Power consumption of greater than 300 watts per GPU (or GPGPU co-processors) are becoming the norm and can easily be addressed with Asetek technology. Similar to Asetek CPU cooling, GPU coolers feature the integration of the pump and cold plate. Depending on the OEM design, these coolers can be in series with CPU coolers for redundancy in node level cooling. In HPC clusters, there is often a demand for nodes that contain only GPUs or GPGPUs. In these cases, GPU cooling follows the same routing of cooling liquid as CPU nodes, but from GPU to GPU cooler rather than CPU to CPU
The supporting VRs and other silicon for these GPUs can also require liquid cooling. Asetek's highly flexible modular cooling system allows this supporting silicon to be liquid cooled as well.
Memory modules are a growing concern as hot spots within servers. Asetek's patent protected memory coolers directly cool server memory with specially designed cooling chambers within the cooling loop. Heat transferring gap pads are used to increase heat capture into liquid and fill the space between the memory and the liquid tubes of the memory cooler for optimal cooling performance without sacrificing serviceability.
A defective memory module can easily be replaced by simply moving the non-sticking gap pad from the defective module to the new module. Memory is sometimes mistakenly looked over when considering hot spots in a server. Although an individual memory module may draw only 5 watts, a 2U 4-Node server might have 64 of these modules for a total of 320 watts. Multiplied across an entire data center, this power load can quickly add up. By removing this heat via liquid instead of air, the cost for to cool the data center is significantly reduced.
Tubes / Quick Connects
Asetek’s tube and connector system is the result of extensive research to optimize flexibility in being able to quickly address differences in server board layouts, assure low permeability, high durability and cost effectiveness for OEMs. Asetek undertook an 18 month engineering effort to develop a rubber-based compound that achieved low permeability requirements with the added durability and flexibility demanded in server applications. The tubing also provides ultra-low permeabitlity that eliminates the need for big reservoirs and concerns during a server’s life cycle for monitoring server fluid levels.
Additionally, Asetek's rubber tubes are less prone to kinking, less likely to be damaged and able to sustain more bends than FEP plastic tubes commonly used. In the data center, reliability is paramount and Asetek's patent pending dripless quick connectors reflect this requirement. Expensive connectors are required by most competing liquid cooling solutions. With Asetek’s distributed pumping architecture, the pressures used on the server side of the cooling system are very low. As a result, very little strain is placed on the joints in the system and failures that can occur in high pressure systems are avoided.
Asetek's Monitoring System is integrated into both InRackCDU and VerticalRackCDU. Asetek engineered software reports real time, rack-level data and is capable of sending alerts (e-mail, SMS, IM, local alarm) at user defined thresholds. Embedded computer hardware and sensors provide a real time view of the RackCDU status.
Asetek's Monitoring System allows direct monitoring of a single rack or clusters of racks via a web browser interface, or integration of the monitored data into a lager DCM (Data Center Monitoring) system via SNMP. In addition, the software is capable of detecting leaks on individual servers and, in the rare chance that a leak should occur, the control software will shut down the pumps in that specific server.
When used in conjunction with Asetek's Monitoring System, RackCDU is capable of reporting the following information:
Radiators (also known as heat exchangers) transfer thermal energy from one medium to another for the purpose of cooling server components and are used for selected OEM solutions.
Asetek In Server Air Conditioning technology (ISAC) targets 100% server heat capture with a combination of D2C cooling and radiators to scavenge the remaining heat from the air stream.
In ISAC, cold liquid enters through the tubes on the back of the server and in to liquid-to-air heat exchangers. In addition to the cold liquid cooling the traditional hot spots (CPUs, GPUs, Memory), the cold air from the liquid to air heat exchangers cools all remaining components inside the server. All air stays and recirculates inside the server and is sealed from the surroundings, eliminating the need for CRAC (Computer Room Air Conditioning) entirely. See more information HERE
Asetek ServerLSL is a “Liquid Enhanced Air” cooling for server nodes which is normally used to integrate the highest performance processors into OEM servers but still use normal data center air cooling for heat rejection. D2C cold plates use hot water to pull the heat from high heat flux components and transfer the heat to radiators inside the node. The heat is then dumped into the air stream.
The radiators used for ISAC and ServerLSL are 40mm, sized in increments that support from 3 to 7 fans, and fit nicely in a 1U server enclosure. Patented modular fan ducts directly clip to the radiator and facilitate fan and chassis mounting. The radiators are braised aluminum and RoHS compliant.