RE: Update on e1000 troubles (over-heating!)

Here is the lspci information, both -x and -vv.  This is with 
two of the e1000 single-port NICS side-by-side.  I have also 
strapped a P-IV CPU fan on top of the two cards to blow some 
air over them....running tests now to see if that actually 
helps anything.  If it does, I'll be sure to send you a picture :)

Ben, I checked the datasheet for the part shown in the lspci dump, and it
shows an operating temperature of 0-55 degrees C.  You said you measured 50
degrees C, so you're within the safe range.  Did the fans help?

Here's the datasheet:
http://www.intel.com/network/connectivity/resources/doc_library/data_sheets/
pro1000mt_sa.pdf

-scott

On Mon, Oct 14, 2002 at 07:20:04PM -0700, Feldman, Scott wrote:

quoted

Here is the lspci information, both -x and -vv.  This is with 
two of the e1000 single-port NICS side-by-side.  I have also 
strapped a P-IV CPU fan on top of the two cards to blow some 
air over them....running tests now to see if that actually 
helps anything.  If it does, I'll be sure to send you a picture :)

Ben, I checked the datasheet for the part shown in the lspci dump, and it
shows an operating temperature of 0-55 degrees C.  You said you measured 50
degrees C, so you're within the safe range.  Did the fans help?

The thermometer he used likely showed a much lower temperature than what was 
actually on the die. 5-10 C more are not unlikely. It's hard to measure chip
temperatures accurately without an on die thermal diode or special kit.
So I would expect that when an external normal thermometer showed 50C 
it was already operating out of spec.

-Andi

At 4:37am +0200 10/15/02, Andi Kleen wrote:

 > Ben, I checked the datasheet for the part shown in the lspci dump, and it

quoted

 shows an operating temperature of 0-55 degrees C.  You said you measured 50
 degrees C, so you're within the safe range.  Did the fans help?

The thermometer he used likely showed a much lower temperature than what was
actually on the die. 5-10 C more are not unlikely. It's hard to measure chip
temperatures accurately without an on die thermal diode or special kit.
So I would expect that when an external normal thermometer showed 50C
it was already operating out of spec.

The datasheet's for the card, so the operating temperature is surely 
ambient, not die temperature. "Ambient measured how?" would be a 
reasonable question, though.
-- 
/Jonathan Lundell.

Feldman, Scott wrote:

quoted

Here is the lspci information, both -x and -vv.  This is with 
two of the e1000 single-port NICS side-by-side.  I have also 
strapped a P-IV CPU fan on top of the two cards to blow some 
air over them....running tests now to see if that actually 
helps anything.  If it does, I'll be sure to send you a picture :)

Ben, I checked the datasheet for the part shown in the lspci dump, and it
shows an operating temperature of 0-55 degrees C.  You said you measured 50
degrees C, so you're within the safe range.  Did the fans help?

Here's the datasheet:
http://www.intel.com/network/connectivity/resources/doc_library/data_sheets/
pro1000mt_sa.pdf

I get some strange e1000 failures too.  It usually involves the 
watchdog kicking them back into order, but sometimes they'll stay 
offline for a while.  Heat would explain it, though, because it only 
happens when I'm actually using the cards for a benchmark.  I figured 
that it was either my cables, or a shoddy switch.

The new dual-port e1000 that I have doesn't seem to have this problem, 
even though I'm running 4 times more traffic than the singles that I 
had.
-- 
Dave Hansen
haveblue@us.ibm.com

Feldman, Scott wrote:

quoted

Here is the lspci information, both -x and -vv.  This is with 
two of the e1000 single-port NICS side-by-side.  I have also 
strapped a P-IV CPU fan on top of the two cards to blow some 
air over them....running tests now to see if that actually 
helps anything.  If it does, I'll be sure to send you a picture :)

Ben, I checked the datasheet for the part shown in the lspci dump, and it
shows an operating temperature of 0-55 degrees C.  You said you measured 50
degrees C, so you're within the safe range.  Did the fans help?

The fan did help, and Andi is right, the chip was much hotter than what
my probe read (I was gently pushing it against the top of the chip, cause it
was too hot to really press my finger against it to get good contact :))

With the fan blowing on the chips, it has been perfect.  This implies to me
that if you are going to run the e1000, you need significant air-flow over
the chipset, and the generic 2U chassis that I have is definately inadequate,
partially because the MB is so big that the fans are too far away from the
PCI slots...  This is all doubly true if you are running two NICs side-by-side,
which is what I was doing.

I am also considering glueing heat-sinks onto the main chip, which may make it
work in more marginal environments.

Ben

-- 
Ben Greear [off-list ref]       <Ben_Greear AT excite.com>
President of Candela Technologies Inc      http://www.candelatech.com
ScryMUD:  http://scry.wanfear.com     http://scry.wanfear.com/~greear

Dave Hansen wrote:

I get some strange e1000 failures too.  It usually involves the watchdog 
kicking them back into order, but sometimes they'll stay offline for a 
while.  Heat would explain it, though, because it only happens when I'm 
actually using the cards for a benchmark.  I figured that it was either 
my cables, or a shoddy switch.

The new dual-port e1000 that I have doesn't seem to have this problem, 
even though I'm running 4 times more traffic than the singles that I had.

That was exactly the behaviour I noticed.  I believe it's because when you
run two side-by-side, they cook each other (I'm assuming you didn't run
2 2-ports side-by-side)

Try strapping a fan on them somehow and I bet all your troubles go
away (and maybe your .ibm email will shame Intel into putting heat-sinks
and/or small fans on their NICs... ;)

(I ran two Netgear 302t NICs (tigon-3) side-by-side for 4 days at max speed, and they
  didn't drop a single packet, even though their heat-sinks were too hot to
  touch!)

Ben

-- 
Ben Greear [off-list ref]       <Ben_Greear AT excite.com>
President of Candela Technologies Inc      http://www.candelatech.com
ScryMUD:  http://scry.wanfear.com     http://scry.wanfear.com/~greear