Operating temperature range for compute module?

[amrbekhit]

Hello,

With the Model A and Model B RPis, the Foundation understandably didn't make any guarantees regarding operating temperature range since integration into commercial/industrial products was not their primary intent.

However, with the impending release of the compute module, aimed at "business and industrial users", will the Foundation be officially publishing temperature specs? The only information available so far has been in the comments section of the product announcement, where Liz quoted James saying that he expects the range to be -25 to +80 based on the chips with the narrowest range. With this information being mentioned in the comments section, I wasn't sure how "stable" that information is and whether it would be subject to change.

Any thoughts?

Thanks

Amr

[jamesh]

Although it's quite feasible to quote temperature/humidity specifications for the individual components, and also the CM as a whole, these numbers are greatly affected by the case in to which the device is put. So the end user will still need to perform their own testing. For example, a nice closed case with a CM in will survive quite nicely at temperatures much lower than the specs as the enclosure will be kept warm by the SoC. And a CM is an actively cooled case will go to much higher environmental temps that the SoC could handle if exposed.

So in what sort of container should the CM be tested? Bare? Which is a situation unlikely to happen in the real world?

[amrbekhit]

Hi James,

Thanks for your response. Yes, I agree with you regarding the case - a board inside an airtight plastic box would benefit from self heating than a board in an air cooled case and thus might have a lower minimum operating temp.

So what I ought to ask is the following: can the Foundation make any official statements regarding which component on the CM has the narrowest temperature range? Or perhaps "which component has the highest MIN TEMP and which component has the lowest MAX TEMP"?

Amr

[Ravenous]

On the minimum temperature case:

The CM, while running in a box, will heat that box. However if it has to start from cold, there won't be any heating benefit. That is the worst case I can think of. (I have no idea if it's fair to assume there won't be any ice in the circuitry in those conditions.)

Even then, you probably have to test some. (I think I dimly remember someone reporting they'd tested a plain raspi after putting it in the freezer.) I don't know how people in the industry determine recommended limits - I'd guess experience.

[elektrknight]

On the minimum temperature case:

(I think I dimly remember someone reporting they'd tested a plain raspi after putting it in the freezer.) I don't know how people in the industry determine recommended limits - I'd guess experience.

If your stuff is big enough to test it you get to take it above arctic circle and then somewhere close to the equator - much better
If it is small like a bus or truck they put it in a strange combination of a deep freezer and a furnace aka environmental test chamber, same with electronics.

You can easily do all this at home, you just need to repurpose hair dryer and blow the hot air over RasPi for a while.
Resist using heat gun as it can melt solder and blow chips of your RasPi PCB.

For testing in low temperatures bag of salted ice will get you close to -15C, just put in a foam container with your RasPi
and you have your test chamber.

[James Adams]

This is what I sent to Liz who posted it in the CM Blog comments:

"Theoretically the operating temp. range is determined by the range of the component with the smallest range.

The Samsung eMMC and Samsung LPDDR2 are all rated for -25 to +80 degrees C, so the range is -25 to +80. (BCM2835 and ana SW are greater range than this, caps are -25 to +85).

However that is the max range for the silicon die – so a user would have to take into account the heat generated when in use and make sure this does not cause the temp. to exceed 80C."

[elektrknight]

However that is the max range for the silicon die – so a user would have to take into account the heat generated when in use and make sure this does not cause the temp. to exceed 80C."

I recall that using "vcgencmd measure_temp" we can get the temperature from a on the die sensor(?).
But beyond that is there any thermal runaway protection built-in the BCM2835?

[jamesh]

However that is the max range for the silicon die – so a user would have to take into account the heat generated when in use and make sure this does not cause the temp. to exceed 80C."

I recall that using "vcgencmd measure_temp" we can get the temperature from a on the die sensor(?).
But beyond that is there any thermal runaway protection built-in the BCM2835?

Yes. When the Soc hits 85 it is clocked back to reduce temperature.

I'm not sure if anyone has ever hit 85. My recent tests on a CM, running the GPU and ARM at full speed, showed 45. Not in a box though.

[tskwara]

Here's an excerpt from our article on overclocking that includes results from our basic thermal testing:

Temperature Tests
For my tests I used an Raspbian image with a pre-built version of Quake3. A zip file containing the image can be found here. For each set of tests I re-flashed my SD card with a fresh image. I used the vcgencmd command to monitor the core temperature of the Broadcom chip and SDRAM. Documentation on the vcgencmd command can be found on the Linux wiki here.

For each round of tests I ran Quake3 on high quality settings and monitored the temperature and clock speed on my laptop over SSH. I started a Quake3 match in spectator mode, so only bots would play until a score of 50 is reached. This usually took about an hour and would keep the ARM CPU at the overclock speed for the entire time. I recorded an average of five temperature checks each hour. The results are as follows:

Code: Select all

Raspi-config 
Setting          Initial Temp   One Hour   Two Hours   Three Hours
None                   39.9°C     42.6°C      47.5°C        49.1°C
Modest                 40.8°C     43.9°C      48.9°C        49.9°C
Medium                 42.8°C     45.5°C      50.2°C        52.0°C
High                   45.2°C     47.2°C      52.5°C        53.5°C
Turbo                  47.1°C     52.0°C      55.6°C        57.5°C
Turbo / Proto Armour   45.5°C     48.5°C      50.2°C        52.6°C

[amrbekhit]

@James Adams. Thanks for the clarification!

[elektrknight]

@tskwara
The overclocking article is quite informative especially the hard data on temperature rise.
Thanks for posting!