It works!

Morgaine, I just did the "finger switching" test again as I suggested. :-)

It seems my LAN chip is a bit hotter than I remember. I too can't get much beyond 5sec wtihout hurting my finger. It could also be that the outside temperature is now 25 degrees, while before it may only have been 20 degrees.

The "can't keep finger on" temperature is pretty much 52 degrees C. No matter who you are or if you say you have fire resistant fingers.

If the outside temperature rises, more heat will need to flow through your finger (less is being diverted by the PCB), and the hotter it gets. Even a 2 degree change might mean two out of two people now report unable to hold beyond 5 sec, while before it was "possible to hold indefinitively"

If you look at the datasheet, on page 41, it says you can expect a typical power use of about 763 mW.

http://www.smsc.com/media/Downloads_Public/Data_Sheets/9512.pdf

I don't pretend to be able to estimate the power-use within  a factor of two, so when I say "about 1W", this is certainly within range.

Regretably the thermal resistance isn't specified. The Texas instruments VCA8500 provides a "typical" value (i.e. a wild guess that it might apply to the lan9512), of 22 degrees / W. At 0.7W, that would mean only about 40 degrees C. We're seeing more than that.

But if the wild guess is off-by-a-factor-of-two, and it's 40 degrees/C for teh lan9512, then 60 degrees sounds like a possible value.

With the case at around sixty degrees, the inside will be only about 80 degrees. This is well below the maximum of 150 degrees.

Chips are built to work at these temperatures. Sure, I too have heard about some chips working themselves loose from thermal stresses. Before you complain to Farnell, the RPF or me, please wait until you KNOW it will fail because it HAS failed.

Some chips use a bit of power, and therefore they dissipate heat. That heat needs to be carried away and a temperature gradient is required before heat starts moving. This is simply the way things work.

Now usually these things are hidden inside a case. So you don't get to poke your fingers at it. The 'pi is open, As a learining experience YOU get to poke your fingers at it. Have you learned something today? That's the stated primary goal for the RPF.

Roger Wolff wrote:

 

Chips are built to work at these temperatures. Sure, I too have heard about some chips working themselves loose from thermal stresses. Before you complain to Farnell, the RPF or me, please wait until you KNOW it will fail because it HAS failed.

This is a technical site.  Questions and discussion are welcomed by Element 14, on both technical and product matters, and your suggested "wait until you KNOW it will fail" is not really helpful advice at all in an engineering community.

We can do a lot better than that, understand what is going on, and determine whether the board is being operated outside of its thermal spec (it's not device survival that's the worry, but the board/joints).  According to Gert, such LAN9512 temperatures were not expected at design time.

The board isn't failing yet, but if it wasn't designed for these thermal stresses then it won't survive them for long.  Element 14's input really ought to be sought regarding what we should do.  I  don't want my board failing on the day after warranty expires when the solder joints finally decide that they've suffered thermal movement long enough, and it's not in the manufacturer's interests for that to happen either.

And once they're in stock and I want to buy several Pi, I need to know that this common operating condition is within board spec before ordering.  (Model A is looking very attractive, the absence of LAN9512 making it more so.)

Morgaine.

I got some of these from SparkFun ...

http://www.sparkfun.com/products/10746

and waiting for these to try on the SoC ...

http://www.ebay.com/itm/ws/eBayISAPI.dll?ViewItem&item=120820119736&ssPageName=ADME:L:OC:US:3160#ht_500wt_1180

-J

Adding a heatsink is certainly an option, but if it's required for board longevity then the heatsink should be part of the board design.

I have no shortage of small heatsinks here, but at this stage it's more a matter of establishing the principle with Element 14.  If they accept that it may be a good idea to add your own heatsink pending review of the situation with board designers, then they may want to advise accordingly since it could affect product warranty.

Agreed.

Also if by any chance, as reported before, due the wrong connection of the LAN9512 VDDCORE pin to the 1V8 rail the GPU is drawing some current from the internal regulator on that chip, it will surelly get hotter than normal.

Ethernet PHY's tend to run warm to hot, but this one seems to be slightly above of what I've seen before.

-J

Really, the chip is using a significant portion of a watt. On a 9mm x 9mm chip that's going to make it feel hot to the touch.

Once I have my "spare" pi, i'll see if I can find that 1.8V line and cut it.....

All measurements so far have indicated a significant powerdraw on the 3.3V line, and that mostly goes to the lan9512. And from there, it can't go anywhere but turn into heat.

There is a post on hackaday.com with someone who has added a heatsink to his 'pi.

That would be something hard to do given that it is two not contiguous pins that probably go to the power plane and you still need to keep the connection to the decoupling capacitors required for the LAN9512.

Unfortunatelly (once again the veil of secrecy) the Gerbers for the board layout are not publically available so it is difficult to tell by just looking at the two pictures of the boards without components how those traces are routed, that btw as far as I can tell are improperly routed since the trace is supposed to go straight from the pin to the capacitor.

I've seen the hackaday post, which motivated me to search for tiny heatsinks to try since I don't want to use the saw :-)

So far in my meter the Rpi shows 380-440mA draw now with the keyboard and tiny mouse connected and a NTSC TV monitor on composite video.

-J