[Server-devel] do RPi 3 B+ still need heat sinks?

[James Cameron]

On Tue, Apr 03, 2018 at 09:07:22AM -0400, Adam Holt wrote:
> On Tue, Apr 3, 2018 at 1:53 AM, James Cameron <[1]quozl at laptop.org> wrote:
> 
>     Temperature readings are not comparable between RPi 3 and RPi 3 B+
>     because of throttling differences, and the metal heat spreader.
> 
>     Ambient temperature is critical for the separate chip used for
>     Ethernet and USB; it has a lower maximum operating temperature than
>     the CPU.
> 
>  
> Do you happen to know this chip's max operating temperature spec?

Only 70 degrees C.

http://www.microchip.com/wwwproducts/ProductCompare/LAN7515/LAN9514

This looks like a binning threshold.

Let me spin a story.  The manufacturer of the chip tests the operating
temperature range using some method, and separates the results into
four bins;

- operates above 86, into the bin labelled "excellent",

- operates above 70 and below 86, into the bin labelled "industrial",

- operates above 60 and below 71, into the bin labelled "commercial",

- doesn't operate above 70, into the bin labelled "throw away".

Then what happens next is often surprising;

1.  if a customer orders the industrial version, it will be filled
from the "industrial" and then the "excellent" bin,

2.  if a customer orders the commercial version, it will be filled
from the "commercial", "industrial" and then the "excellent" bins.

The chips can be encapsulated and labelled at any time, before or
after this binning.

So you as customer of customer only knows it will operate to 70, and
it might operate to 85.

The chip heats from being powered, and from Ethernet and USB activity.

The heat from the CPU will bleed across to the chip.  In a normal
enclosure, this isn't a problem.  In a sealed enclosure with no air
flow, and high CPU, Ethernet and USB activity, it could be more
interesting.

It would be reasonably easy to test for; if the CPU temperature does
not fall rapidly when activity slows, it is because the board and
air nearby has reached equilibrium with the CPU.

>     Your best comparison will be between timings of equal workloads with
>     equal cooling in identical environment.
> 
>     You can't do this honestly with temperature alone.
>    
>     --
>     James Cameron
>     [2]http://quozl.netrek.org/
>     _______________________________________________
>     Server-devel mailing list
>     [3]Server-devel at lists.laptop.org
>     [4]http://lists.laptop.org/listinfo/server-devel
> 
>     --
>     [5]
>     [6]Unsung Heroes of OLPC, interviewed live @ [7]http://unleashkids.org !
> 
> References:
> 
> [1] mailto:quozl at laptop.org
> [2] http://quozl.netrek.org/
> [3] mailto:Server-devel at lists.laptop.org
> [4] http://lists.laptop.org/listinfo/server-devel
> [5] http://lists.laptop.org/listinfo/server-devel
> [6] http://lists.laptop.org/listinfo/server-devel
> [7] http://unleashkids.org/

> _______________________________________________
> Server-devel mailing list
> Server-devel at lists.laptop.org
> http://lists.laptop.org/listinfo/server-devel


-- 
James Cameron
http://quozl.netrek.org/