there was a mini-symposium held on may 30, in conjunction with itherm in san diego. the symposium was well attened even though it was from 7 to 9 pm. the presentations were mainly by vendors who were describing their liquid cooling solutions. i personally enjoyed the technical presentation by kathryn whitenack of lytron. she provided very useful information on cost drivers for a liquid cooling system without any attempt to sell lytron's products.
it seemed that combining the pump and the fan was a popular strategy for reducing the size of the loop. intel is developing a solution where the pump is embedded in the cold plate.
the main problem in comparing the performance of these systems was the lack of a consistent set of performance curves with similar units. the vendors seemed to avoid showing such performance curves and focused on case specific data.
in a liquid cooling loop there are 3 almost universal numbers. these are the numbers you start with when you are looking for a solution to your thermal management problem.the first number is the ambient temperature where the heat is being rejected to. then we have the heat flux. i say heat flux because just the power dissipation is not enough. you need to know the surface area over which the power is applied. the third number is the surface temperature of the cold plate that is in contact with the package or the interface material. you normally start with the maximum allowed temperature for thie die, account for the temperature rise through the package and the interface material and arrive at a desired cold plate temperature. if you divide the difference between the surface temperature and the ambient temperature by the heat flux we get a number that we call loop thermal resistivity. if you plot this number versus the flow rate per unit area of the heated surface, we will have a nice and application independent performance curve. the total loop performance depends on the performance of the cold plate, the radiator (liquid-to-air heat exchanges) and the pump. at the end of the day, however, it is the loop performance that you care about.
the thermal performance, however, is not the only criteria you use for choosing a solution. other issues such as reliability, cost, vendor reputation, etc. come into play. but, with a unified reporting of the performance curves, you know what solution is in the game, to begin with.
the only way that this happens is for the end-users to demand such technical data. most vendors will not voluntarily publish such data as it makes the comparison of various solutions very straighforward and may take them out of the race quickly.
|