Ethernet chip gets raging hot

[spurious]

Yes FOV taken into account..
Just to satistfy the weird people who think fingers are thermomters I can held my finger on the new board I got from Farnell.
IR thememeter
Ambient 25.1C
CPU 48.9C
Eth 50.8

These values are quite different from what you reported earlier...

Best regards
Simon

[simonbr]

There IS a heat problem if the chip is meant to be running at or below 30C.

According to the data sheet, its operating temperature specification is up to 70 degC (assuming the commercial type is mounted). I have the impression that as reported in this topic, the chip doesn't get as hot as that. I wonder though how much more the chip heats up under prolonged heavy internet traffic.

Best regards
Simon

[simonbr]

Yes FOV taken into account..
Just to satistfy the weird people who think fingers are thermomters I can held my finger on the new board I got from Farnell.
IR thememeter
Ambient 25.1C
CPU 48.9C
Eth 50.8

These values are quite different from what you reported earlier...

Best regards
Simon

I had noticed that part, but one way or the other I find it remarkable that the temperature differences from ambient you reported now are up to 2.5 times as high as those measured before. Would there be so much difference between two boards???

Best regards
Simon

[HenryG]

I can barely stand the heat on the controller chip on my PI, with just the keyboard attached. My finger nearly hurts. It's around 24ºC ambient here.

[Yoda007]

I am worried about the temperature too. Got my unit yesterday, and i noticed that if i put my finger over the SoC i can feel the heat (with 2% cpu load (the process top itself)).

I refrained from touching the actual chip, don't know if that could short it or something.

Is that normal ?
There probably isn't a temperature sensor on the chip so i can check the actual temperature ?
How long is the warranty (It works now, but what if it dies because of the heat)?

[spurious]

chips can get hotter than finger comfy.. fact
different manufacturing can produce diffident results.. fact
hot chip from finger thermometer does not mean faulty.. just means wrong method of measurement.. fact
bored of this thread now.. go read about the chips and what their thermal properties are.. then come back here with results.. you can Google too!!!

[mahjongg]

agreeing with spurious. If people would poke around in other electronics stuff they would get blisters!

The PI needs a case, just so to prevent people going "hey this chip is hot, so something must be wrong".

And for what its worth, I think that case doesn't have to have ventilation holes!
The whole darn thing uses just 2 Watt! quod suus omnes

[Yoda007]

Thanks, I thought I was paranoid. Happy and unhappy to find out its true .

[iwant5please]

Here is some useful information for anyone who thinks their Raspberry Pi is getting too hot: http://www.youtube.com/watch?v=mjH402mca8E

[milhouse]

Here is some useful information for anyone who thinks their Raspberry Pi is getting too hot: http://www.youtube.com/watch?v=mjH402mca8E

Nobody knows what the ""normal operating temperature" for a Pi is - you can't determine these metrics from a single sample, which is why it needs input by someone from the Foundation that understands the hardware design.

Nobody is suggesting the Pis are overheating, just wondering if some devices might be running hotter than intended (based on vague descriptions of them supposedly running "cool" and Gert concurring that one device was running "too hot").

[rurwin]

There isn't a single sample. There are several people now agreeing, by one means or another, that the normal operating temperature is around 50C.

I've got my Pi in the meantime, and my calibrated finger agrees: hotter than a hot bath, but not too hot to touch. 50C.

[jchadwick]

My SMSC chip is heating up when WiFi is enabled (via USB WiFi.) Heat does not stop spreading: I had it running while I was gone for a while and it ended up overheating my wireless adapter.

This is just with idling...

So, what should I do? Any way to "slow" WiFi/USB/etc.? Best off with a heatsink?

[Mega]

It is definitely not a heat-related issue, unless you genuinely have a defective unit. I used a fan to cool the smsc(network/usb) chip in question and I was able to reliably crash the Pi despite the surface of the chip being cool to the touch. I have fixed(for now; I will not rule out a hardware defect) the problem on my Pi.

For anyone still having this issue, please try one of these two solutions(I recommend solution 1):

Solution 1: In your /etc/sysctl.conf, put this at the bottom:

Code: Select all

vm.min_free_kbytes = 8192

This allows smsc95xx turbo mode to have enough memory to do its thing without overflowing kernel mem and causing the crash. You may set this higher if you still have the problem, until it goes away. Don't set it too high though as this will restrict more memory from programs(though, not a lot; this solution reserves 8MB for the kernel).

Solution 2: In your /boot/cmdline.txt, put this at the beginning:

Code: Select all

smsc95xx.turbo_mode=N

This disables the smsc95xx turbo mode(as far as I can tell, I don't know why the value must be N) which, when enabled, allows the network driver to send multiple frames at the same time. This can overflow the kernel memory and cause the crash. Disabling turbo mode may reduce network performance. This could be an acceptable tradeoff for some people.

I found that I could reliably crash my Pi by checking out the svn trunk of ioQuake3. To test this, cd into /tmp, and run this command(make sure you apt-get install subversion first):

Code: Select all

svn co svn://svn.icculus.org/quake3/trunk ioquake3

Try this test before and after you attempt one of the solutions above. If it crashes afterward, try the other solution. If they both fail to fix the problem, then you have another problem. This information comes from a bug on the Linaro Linux section of Launchpad: https://bugs.launchpad.net/linux-linaro/+bug/664477

I prefer to set the minimum kernel memory to 8192 so I can keep turbo mode, as I plan on doing a lot of network streaming with my Pi. I hope I helped a bit here. If I got any terminology or other information wrong, please correct me. I want to become more knowledgeable in this.

[Ray_GTI-R]

Hi. Sorry to butt in. Completely independently of this and similar threads I found a hot spot around the 6 tiny components near the RJ45 socket.
My solution:-
Pic 1 my "case", with 2 x 2mm (details elsewhere) thermal gap filler*

Pic 2 the "case" upgraded with a 1mm copper plate and the RJ45 heatsink

Pic 3 a close-up of the RJ45 heatsink & gap filler.

Some details. All made from stuff I had lying around, no special measurements were taken. The heatsink is half of a de-anodised (was black) Maplin micro sink. There is a 1mm strip of thermal gap filler covering the 6 tiny onboard components - and the solder pads of the nearby takeoff socket to avoid shorting out on the heatsink. The heatsink squashed down onto the onboard components then stuck to the RJ45 socket using Sekisui #5760 - it won't move.
Using the little 25mm x 10mm USB-powered fan (see Pic 2) the temperatures are now only exceeded by the two main chips which thanks to the little fan show only 2 or 3 degrees above ambient.
Job's a good 'un.
Ray
(*Excuse the killer mounting system - two rubber bands - which keep the 'board pressed down firmly onto the 2 x 2mm gap filler ... must spend some time on this soon! )

[jbeale]

Interesting mounting project! I don't think that amount of cooling is needed, but it looks cool nonetheless . The tiny components near RJ-45 socket might be the 50 ohm ethernet terminating load resistors R22, R23, R24, R28. They will dissipate some heat whenever the network is active. 100baseT Ethernet uses +/- 1 V for signalling, but those resistors are connected to +3.3V. If there is 2.3 V across a 50 ohm resistor, that is P = V^2/R = 100 mW per resistor. Maybe only half that, if you have a 50% duty cycle, or 50 mW. That is almost the limit for a standard 0402 size resistor (63 mW) so I expect it would get hot.

[Mega]

A heatsink system is unnecessary for the Pi... just set your sysctl.conf right and you won't have any problems.

[Ray_GTI-R]

Hi.
Hehe - the fundamental idea for the "case" was that I got fed up of the RPi wobbling about if I sneezed - now it doesn't.
Heatsoak through the 'board (any board) has been around since the year dot. Now there's none.
I could be wrong but as I understand passive, non-logic components like resistors, sockets, VRMs etc etc they aren't directly controlled by software. In my standard version of /etc/sysctl.conf - which has three executable lines, two appear to be RPi-specific (swappiness & free_kbytes) - I can't see anything to help cool off the RJ45 socket/ethernet terminating load resistors

[whiteshepherd]

I'm having a serious crash problem with my pi (RS) when it gets hot. I "thought" was related to heat (might be?).

Here is my setup and symptoms and hopefully someone can tell me some things to try?

We live up on the US/Canada border and it doesn't normally get hot in the house (shaded by trees, etc). However this year has been extra hot and the temps in the home have got to the upper 80sF on some days in the house.

My Pi is hooked up on a desk with no USB connected. Only Ethernet is plugged in. Network traffic is light but constant. I have my hobby IRCD server compiled and running well on the Pi and 25-50 kids chatting up a storm 24/7 non stop all day long. I have a plastic box over the PI to prevent dust. Two walls of the box (right & left) are removed for ventilation and the box does not physically touch the Pi except for the USB port which it almost touches on top. The Ethernet port is plugged directly into my Tomato WRT54GL.

Like I said when it gets hot the Pi will crash. I know it crashes as I will see all the lights flashing happily along then suddenly all the lights go out except the red power light. When I check the chips, Ethernet port, and the USB port are all WAY to hot to touch. So I assumed it was a heat issue. If I unplug power and plug it back in only the reg power light comes on and the Pi does not boot. If I unplug the power and wait 3 min (to cool) then plug in the power the Pi boots just fine with no problems until it overheats again (if it is overheating). MY Pi is not overclocked and running at factory default specs.

From this thread I checked /etc/sysctl.conf and it already had vm.min_free_kbytes = 8192 in there. Default?

I know this is planned for poor children in third world countries. Many of them do not have any AC if this is a heat issue?

Do you think my issue is heat or something else? What are some things I could check or try?

[johnbeetem]

I'm having a serious crash problem with my pi (RS) when it gets hot. I "thought" was related to heat (might be?).

Here is my setup and symptoms and hopefully someone can tell me some things to try?

...

Do you think my issue is heat or something else? What are some things I could check or try?

It sure sounds like a heat problem. I found that mounting my RasPi vertically reduces her temperature since a lot of heat gets conducted to the bottom side of the board. By mounting her vertically, she gets natural convection cooling on both sides. Adding a heat sink to the LAN9512 might help as well, since that's the hottest component on the board.

There's a marvelous analysis by Remy of RasPi heating with color temperature images here: http://www.geektopia.es/es/technology/2 ... iento.html ["Is the RasPi computer getting hot? A study of its operational temperatures", in Spanish]

[Mega]

I can't see anything to help cool off the RJ45 socket/ethernet terminating load resistors

You don't need to... they are designed to operate normally at these temperatures.

I'm having a serious crash problem with my pi (RS) when it gets hot. I "thought" was related to heat (might be?).
...
Do you think my issue is heat or something else? What are some things I could check or try?

I doubt it is heat-related.

From this thread I checked /etc/sysctl.conf and it already had vm.min_free_kbytes = 8192 in there. Default?

Try changing it to 12288. I have found that with even heavier network load that it can still crash with 8192.

[janisalnis]

I got a problem that RPi hangs during intensive network traffic.
Chips get very hot inside enclosure. I put a small fan. Got better. But did not solve problem.

With command "free" see that RAM gets filled up when coping large files using wget.

total used free shared buffers cached
Mem: 190836 179136 11700 0 8528 142832
-/+ buffers/cache: 27776 163060

And it does not get freed after copy finishes.

Helped to set larger reserved free RAM

nano /etc/sysctl.conf
vm.min_free_kbytes = 8192
change to
vm.min_free_kbytes = 16384

[Ray_GTI-R]

Excellent info at http://www.geektopia.es/es/technology/2 ... iento.html if I read it correctly the main chip maxxed out at 65.1C. If that's not hot then I don't know what is ... mine never gets anywhere near that hot BTW.
I spent the first week with my RPi sorting out a crashing issue - it wasn't heat, turned out to be a shoddy phone charger, got the micro-UPS setup with a proper 2 amp plug and the RPi has been rock solid ever since.
HTH, Ray
(Plea to all & sundry ... please don't pluck comments out of context, wrongly, then denigrate them, it just looks silly.)

[computerpie]

I wanted to add an unexpected result I got when I was trying to setup a small fan to cool my raspi.

My usb/eth chips has always ran raging hot, by raging hot I mean you couldn't keep your finger on it for more than a minute. With Raspbian it ran even hotter than with Squeeze. So rather than pasting a heat sink on it I thought I'd setup a way to run a small fan on my board to keep it cool. The fans I have are all 12v and I no longer have any bench supplies so I thought I'd setup a pc power supply and get 5v for the pi, 12v for a fan and projects plus I have 3.3v and -12v available too.

So I took a pc power supply (one that has SATA power cables) and built a circuit board to bring all the voltages to a terminal strip so that I'd have easy access to connect circuits. I'm running my raspi off the 5v from the power supply which is rated for 11A and had a small fan running off the 12v rail.
Voltage at TP1 and TP2 is sitting at 5.1 v and is solid. I'd reach over and touch the chip and noticed it was really cool to the touch. I didn't expect just a fan to make that big of a difference so I unplugged the fan. Without the fan the chip does heat up a little bit, but a noticeable difference when I was running off a phone charger.

That chip has never stayed this cool with an eth cable attached. I have a browser open and have been surfing the net for a couple hours now and I also have a vnc session open to the raspi on my linux box.

This tells me that having a really good power supply makes a difference to how cool the Raspi runs. By good I mean not just one that stays at the suggested voltage when running. My phone charger stayed at 4.9 from start up, but it only had a 700mA rating.

I'm not suggesting everyone run out and hook up a pc power supply, but using this setup and not a phone charger has made a noticeable difference.

Now I need a 7.5v regulator circuit to run this junk Usb Hub.

[jojopi]

My usb/eth chips has always ran raging hot, by raging hot I mean you couldn't keep your finger on it for more than a minute.

Me too. Except that I would call that "a bit warm" rather than "raging hot". You are describing at most 45°C, which is tens of degrees below the point where silicon chips start to see any reduction in lifetime, let alone overheating failures.

The LAN9512 is powered from the 3V3 rail and has no connection at all to the 5V input. So I do not believe that the quality of the 5V supply can have any effect on its power dissipation.

[Mega]

Judging a chip's safe running temperature by how long it takes to hurt your finger is not the best method at all. The chips are designed to run normally at temperatures like this indefinitely.