Copper vs. Aluminum - Thermal Conductivity & Radiation

[Designer]

Remember that discussion about how heat can only flow from a higher temperature source to a lower temperature sink. And also, the heat starts at the circuit junction level and continues to the outside air where it is finally rejected.
Every time there is a heat transfer from one source to another sink there is a delta T. The first sink becomes the source for transfer to the next sink and another delta T is created. The delta T's are additive and the result is the junction temp must increase to "drive" the heat out of the system. More on this below.
But, you CAN release heat to a higher temperature sink by adding work to the system. This is the vapor compression (VC) refrigeration sysrem. The problem with using VC is the power usage effectiveness (PUE) of a data center gets worse. The ideal PUE is 1.0. Meaning all of the electric going into a data center is used for computing. No electric is needed for heat removal. 10 years ago data centers were designed with an "acceptable PUE" of 2.0. Every kw devoted to computing required another kw for heat removal. Factored into that kw for heat removal was the extra energy to operate an N+1 and N+2 equipment strategy for a Tier 4 facility.
Today, designers are being pressured to achieve better overall effectiveness and still maintain high reliability standards. Today, facility designers, server designers, chip designers, and HVAC designers are now working together on the efficient heat removal problem.
Back to the delta T discussion. A heat removal system (with no added VC work) that can reject the junction heat directly to the outside air will minimize the overall delta T and improve effectiveness. Also, a heat removal system that utilizes the phase change phenomena in the fluid will also lower the overall delta T and thus improve effectiveness. I attached a link that just came in from the trade journals.
Interesting Stuff.
Thanks for the Great Conversation.

https://www.nextbigfuture.com/2017/...-of-the-power-of-traditional-air-cooling.html

 

[FordGT90Concept]

That was my thoughts exactly when saying lower power draw. Water cooling introduces more BTUs to the environment. Any system that adequately cools with the fewest resources necessary to complete that task is overall better for the entire system. Efficiency > Power in virtually all things. If you can make a care almost completely dragless, you can make it go a lot faster with a lot less power. It makes sense all around. This is why I'm generally opposed to phase change and water cooling. Instead of dealing with the system heat and the few watts a fan consumes, you're talking exponentially more power consumption which translates to the room the system is in getting a lot hotter which translates to more environment cooling costs which people often overlook. Nevermind more points of failure and vastly greater risk of damage should a cooling component fail. Case in point: friend of mine has a Scythe Mugen on his i7-2600. Fan died a long time ago. Didn't matter because the heatsink is fully capable of cooling his processor without a fan at all. Cooling system that costs nothing operationally. I'm surprised there isn't more emphasis on that in case and heatsink design. Case in point: look at the cylindrical Mac Pro. It has one fan to cool everything and it does it by doing what comes naturally: draw cool air from the bottom of the case and exhaust it at the top. I think it's long past time to consider a new standard for cases that focuses on passive cooling (orient all heatsinks vertical in the case) with optional active cooling (for high power components). From a cooling perspective: ATX doesn't make sense. BTX tried to fix that but I'd argue it didn't go far enough so the standard was largely ignored outside of mass produced machines.

 

[Designer]

Here's a tongue in cheek thought that builds on your recommendation. Once the servers are mounted in the racks, maybe the need for the case can go away. Maybe the server manufacturer can work with the chip, rack, and HVAC manufacturers to build a totally integrated box that has super computing capability with well engineered heat removal.
Wait a minute? We already had that! It was the IBM and Amdahl mainframes of the 60's thru 80's.
The pendulum might be swinging back