"An inferno on the head of a pin” (blog on 1U cooling stress testing)

[Mira Yurizaki]

https://blog.codinghorror.com/an-inferno-on-the-head-of-a-pin/

 

This interesting blog popped up on my news feed. It talks about selecting an appropriate CPU for a 1U server and testing the best way to cool it off while running mprime.

 

The short of it is the author first started with an aluminum heat sink with the included thermal paste. Then switched to copper with the included thermal paste and still found the server overheating. Switching the paste delayed overheating some more. Then finally they built a makeshift airduct that kept the CPU stable.

 

Though a final takeaway was since it was a web server and not a compute machine, take note of the program you're using to thermal stress test the system, since it may not actually be a representative of your use case.

[ARikozuM]

I'm still waiting for those Dell/HP air ducts to be commonplace or at least available. 

[Mira Yurizaki]

Another take away which I forgot to say was Intel's TDP rating can be way off, depending on what you consider a "normal" workload. The 6-core Xeon the author tested dumped 40+ more watts than the 4-core it seemed. If you take power consumption = TDP,  then it grossly exceeded Intel's number

[ARikozuM]

2 hours ago, M.Yurizaki said:

If you take power consumption = TDP,  then it grossly exceeded Intel's number

The chips took more wattage than they were rated for? Or they released more heat than they were rated for?

[Mira Yurizaki]

1 minute ago, ARikozuM said:

The chips took more wattage than they were rated for? Or they released more heat than they were rated for?

If you want to believe TDP = power consumption, then yes to both questions. But I don't believe TDP = power consumption.

[ARikozuM]

1 minute ago, M.Yurizaki said:

If you want to believe TDP = power consumption, then yes to both questions. But I don't believe TDP = power consumption.

TDP is the amount of wattage (heat) that the object was designed to take or safely dissipate. Wattage can be higher than the TDP since not all of it is converted to heat, though most of it is.

[Mira Yurizaki]

15 minutes ago, ARikozuM said:

TDP is the amount of wattage (heat) that the object was designed to take or safely dissipate. Wattage can be higher than the TDP since not all of it is converted to heat, though most of it is.

The problem I'm finding is people think it's the same thing (because watts is watts amirite?) and the industry isn't helping.  It also doesn't help that when you have resistance, the amount of power it consumes is dissipated as heat.

 

Either way, the caveat is don't trust the manufacturer to know what your "typical workload" is and that the TDP is just a blanket one.

[ARikozuM]

15 minutes ago, M.Yurizaki said:

The problem I'm finding is people think it's the same thing (because watts is watts amirite?) and the industry isn't helping.  It also doesn't help that when you have resistance, the amount of power it consumes is dissipated as heat.

 

Either way, the caveat is don't trust the manufacturer to know what your "typical workload" is and that the TDP is just a blanket one.

All energy gets converted into another form. Some of the energy stays in the system due to efficiency, but wasted electricity converts into heat as it's the only way (I'm sure there are more avenues) to get rid of it.