The Fluid Network – Optimizing Hydraulic and Thermal Performance

Webinar

Overview:
Although cold plates are a key component to a liquid cooled system, the fluid network also consists of pumps, heat exchangers, control instrumentation, quick disconnects, fittings, and tubing; all of which require careful planning and engineering. This presentation and discussion offer insights on how to plan and engineer a robust, reliable, and affordable liquid installation by optimizing both the subcomponents and the system as a whole.

Who Should Attend: Program Managers, System Engineers & Component Engineers

Speaker: Chris Chapman

Speaker: Chris Chapman

Chris Chapman has worked in the electronics cooling industry for over 25 years and has been a key contributor for Aavid with many strategic customer programs. Currently Chris is the Director of Product Management, Two Phase, and Liquid Cooling Products.

Below are the audience questions asked during the live presentation.
Click on the question to view the answer.

How could coolant leakage be detected for closed loop systems? Does it need to in-situ measurement?

Using Leak Wires can detect leaks and then could signal shut-off valves to stop the flow. This can be implemented for close loop systems.

How can a fluid to air heat exchanger be used on a larger data center rack to cool the ambient air inside it

In this case, the A2L HEX will be used to transfer the heat from the air to the liquid. Of course, the air temperature needs to be higher than the fluid temperature. This is done frequently.

You mentioned the MTBF of the pump as a concern. Usually this is seals and bearings. What other types of failures have you seen?

The major root cause failures in pumps tend to be with particle size affects in the bearing along with capacitor failure.

I find that a cold plate with ~0.5 mm fin gap has a very small C/W change with change in fluid lpm. A similar analogy with air cooling heat exchangers shows a much greater dependence for airflow vs. C/W. Any physics based explanation for why?

Looking at pure heat transfer coefficients (same for both air and liquid), they reduce as the flow length continues due to boundary layer build up. By increasing the height and the number of fins along with breaking the flow length all will improve the heat transfer coefficient. However, after that there is a limit of the technology. So this is why on the cold plate slide I show different technologies with “h” limits.

For the delta T is that water rise through the block?

The delta T is from the cold plate to the inlet fluid temperature. It is not the fluid temperature rise.

Does Aavid have a line of 2-phase pumps suitable for electronics cooling?

Aavid does not currently make our own pumps. We work with several key suppliers of pumps including pumps specifically developed for 2 Phase.

Is corrosion a low risk or still needs attention?

Corrosion can occur when using materials that are not compatible with the coolant throughout the life of the product. Some coolants due include corrosion inhibitors. This is all analyzed during the reliabilty analysis for the fluid network.

With desionize water what the evolution of performances

Purifying the water affects the electrical resistivity, electrical conductiivty and pH however we do not see any appreciable affect on Specific Heat, Density or Viscosity and so we see no affect on thermal or hydraulic performance.

Follow up, the question was, what fails besides the pump?

In a fluid network, the next failure point after the pump would be a fluid connection point, such as a coupling, fitting or hose joint fatiguing. This is why we recommend fatigue testing 100% of the assemblies. Additionally, corrosion or biofilm growth can also occur if improper materials and coolants are selected.

For He leak test , what is a reasonable criteria for a liquid cooling cold plate loop?

We recommend that 10^-6 or 10^-7 should be used for He Leak testing.

So we expect ~2 liters per min per CPU loop 10C rise allows 77KW / 36 systems ~2KW per loop. 2 CPU/GPEs maybe 200-300W each would you need ~8 cold plates blocks in series to get that level of rise? Is that napkin math correct? Same Pressure assumption?

Because of this Delta T, we need to look at the fluid temperature rise in order to validate the calculations. This is very critical in determining a cooling loops solutions consisting of cold plates in parallel and series achieving proper case temperatures.

Great presentation! How much work is going on with respect to shape optimization for liquid cooling devices?

We analyze and develop our cooling solutions for the devices/interconnects we are cooling for any new or innovating form factor that we have not previously optimized. This is a big focus for us.