Heat sink adhesive

[joan]

The 5.1V figure is probably a misunderstanding. There is no change to the FAQ. It's probably being quoted because the official PSU was from a job lot spec'd at 5.1V.

[masa-aud]

reply to midhun on heat sink
I see this heat issue invites many case studies in specific that are useful for us including problematic adhesives.
I also think that there are not simple methods of heat flow analysis in order to estimate enough cooling tools in need in relation to casing and to the cpu's power consumption characteristics particularly when fully accelerated active conditions etc.. So I think you are welcome to join the experimental study that solving such heat issue of Pi with your adhesive trial e.g..

[Burngate]

The way I understand it, the SoC works on 3v3, and the on-board 3v3 regulator would be happy with anything above that (up to well above 5v, but that's beside the point)
What's not happy with less than 5v is all the USB stuff like keyboards, etc, and possibly the HDMI
The low-power detector switches off the red LED when 5v is low, but won't do anything else.

5.1v or above is recommended for a psu because most free leads are crap, not because the Pi needs it.
A bit like your car needs 200 horse-power because your brakes will probably be binding.

[MarkHaysHarris777]

5.1v or above is recommended for a psu because most free leads are crap, not because the Pi needs it.
A bit like your car needs 200 horse-power because your brakes will probably be binding.

I think that quote ( the one up there ) is quite possibly the worst [ computer <--> automobile ] analogy (fallacy) I have ever heard--- ever. (one for the books)

[Burngate]

As my teacher wrote on my report: "Andrew is trying. Very trying"

[Heater]

So hard to find non-misleading information on the forum....

The ARM/GPU core runs at 1.2 volts. Unless you tweek over_voltage. See here:
http://elinux.org/RPiconfig#Overclocking_options

So I seriously don't believe plus or minus 0.1 volts on the 5v supply makes a jot of difference. It may make issues for attached USB devices but I don't have any except a WIFI dongle.

Now it looks as if I have burned my Pi. xhpl only ever hangs the whole system now.

I have not done any over clocking or over volting. I can't find a heat sink immediately.

What to do?

[Rive]

So hard to find non-misleading information on the forum....

The ARM/GPU core runs at 1.2 volts. Unless you tweek over_voltage. See here:
http://elinux.org/RPiconfig#Overclocking_options

So I seriously don't believe plus or minus 0.1 volts on the 5v supply makes a jot of difference. It may make issues for attached USB devices but I don't have any except a WIFI dongle.

Now it looks as if I have burned my Pi. xhpl only ever hangs the whole system now.

I have not done any over clocking or over volting. I can't find a heat sink immediately.

What to do?

Hate to break it to you, but the page you linked/are quoting for power consumption is for the Pi/Pi2 (not Pi3). Did you not see the Pi image on bottom? or date last updated?

A clue also might have been that the highest Arm freq overclock it goes to is 1150 MHz?

This page was last modified on 23 February 2016, at 02:37.

[Rive]

Now it looks as if I have burned my Pi. xhpl only ever hangs the whole system now.

I have not done any over clocking or over volting. I can't find a heat sink immediately.

What to do?

Also, to stop the benchmark from crashing or failing try adding this (if your power supply will allow it)

/config.txt

Code: Select all

over_voltage=1

[MarkHaysHarris777]

I have not done any over clocking or over volting. I can't find a heat sink immediately.
What to do?

Well, you could have put heat sinks on that thing in the first place rather than arguing with me and Rive. I can't believe you did what he told you to (without heat sinks).



PS... ask Rive about the new water cooling tower he's working on

[Heater]

Rive,

Go ahead, break it to me, I can handle it. You haven't actually said anything there.

I'm confused. I did not make any comment about power consumption, only voltages.

Sure that document may be out of date. If you can link us to anything newer that says differently that would be great.

As it stands I now have an unreliable Pi 3. What to do?

[Rive]

Rive,

Go ahead, break it to me, I can handle it. You haven't actually said anything there.

I'm confused. I did not make any comment about power consumption, only voltages.

Sure that document may be out of date. If you can link us to anything newer that says differently that would be great.

As it stands I now have an unreliable Pi 3. What to do?

You realize the watts (which is voltage x amp) is what allows the power consumption of the Pi to occur?

I already linked an up to date article from the manufacturer's site (element14) regarding the Pi3 power consumption (voltage/amps) at idle.

[Heater]

MarkHaysHarris777,

...you could have put heat sinks on that thing in the first place rather than arguing with me and Rive.

Yep, could have. If I had one around.

I can't believe you did what he told you to (without heat sinks).

It does seem a bit silly now doesn't it?

On the other hand there are no big red warnings on the web site or in the instructions that came with the Pi that a heat sink is required.

Also have read a lot around here about how the SoC measures it's own temperature and throttles back when getting too hot.

Not to mention the assurances I read here from somebody at Pi towers that much testing had been done.

So, who would then expect a piece of software to cause permanent damage?

Let's just say it was an experiment I had to do for myself.

[Rive]

...So, who would then expect a piece of software to cause permanent damage?

The software won't damage the Pi/SoC....The Pi3 is enabled for NEON (which i think is what Linpack bench uses? I know Cpuburn-a53 does).

(e.g., "features: ...neon")

Code: Select all

pi@raspberrypi:~ $ cat /proc/cpuinfo
processor   : 0
model name   : ARMv7 Processor rev 4 (v7l)
BogoMIPS   : 38.40
Features   : half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm crc32 
CPU implementer   : 0x41
CPU architecture: 7
CPU variant   : 0x0
CPU part   : 0xd03
CPU revision   : 4

processor   : 1
model name   : ARMv7 Processor rev 4 (v7l)
BogoMIPS   : 38.40
Features   : half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm crc32 
CPU implementer   : 0x41
CPU architecture: 7
CPU variant   : 0x0
CPU part   : 0xd03
CPU revision   : 4

processor   : 2
model name   : ARMv7 Processor rev 4 (v7l)
BogoMIPS   : 38.40
Features   : half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm crc32 
CPU implementer   : 0x41
CPU architecture: 7
CPU variant   : 0x0
CPU part   : 0xd03
CPU revision   : 4

processor   : 3
model name   : ARMv7 Processor rev 4 (v7l)
BogoMIPS   : 38.40
Features   : half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm crc32 
CPU implementer   : 0x41
CPU architecture: 7
CPU variant   : 0x0
CPU part   : 0xd03
CPU revision   : 4

Hardware   : BCM2709
Revision   : a02082
Serial      : 00000000xxxxxxxx
DNPNWO

Many have successfully linpack benched a Pi without heatsinks (and without damage), and run cpuburn-a53 without damage as well.

What can damage the Pi is excessive heat (if for some reason the pi cannot throttle or otherwise resolve the thermal condition). Common sense dictates if you are running a bench you don't walkaway ,and leave a crashed device running (IMO only a dumb@ss would do that ). In the event of a crash (lock up/unresponsiveness), you IMMEDIATELY reboot the pi (alt+sysrq+b), or failing that and if you don't care about corruption, remove the power source and perform a "hard reboot".

[Heater]

I agree. Seems I was not quick enough off the mark yanking the power when I noticed it had locked up. We are talking seconds here.

There is no rebooting. Nothing is responsive. Only cutting the power.

Surely this is a bug. Either the SoC's temperature monitoring and throttling is not working. Or the documentation needs a big red warning about needing a heat sink.

[Rive]

I agree. Seems I was not quick enough off the mark yanking the power when I noticed it had locked up. We are talking seconds here.

There is no rebooting. Nothing is responsive. Only cutting the power.

Surely this is a bug. Either the SoC's temperature monitoring and throttling is not working. Or the documentation needs a big red warning about needing a heat sink.

You are most likely correct. Seconds are nothing in a crash. Highly unlikely the bench or cpuburn-a53 contributed to any kind of damage...I have read some threads where they left their pi like that for like an hour , thinking all was ok, and wondering how long it was gonna take...


I still think your [. xhpl] issue can be resolved with a proper power supply (and maybe "over_voltage=1" in /config.txt), but a heatsink won't hurt either, I suspect.

[Heater]

Rive,

Thanks. Pi 3 is back to life!

I tried over_voltage 1, 2 and 3. All worked, 0 still does not, so I settled on 2 in case 1 is marginal.

The thing throttles back at about 62 degrees and maxed out at about 78 degrees. So I'm revising my opinion:

Heat sink not required unless you really need all the performance for any length of time.

This is a "proper" power supply by the way. It's about the biggest power bank I have ever seen and is capable of putting out 3 amps without drooping.

Results here:

Code: Select all

...
- Computational tests pass if scaled residuals are less than                16.0
...
frequency(45)=1200000000
temp=58.0'C
frequency(45)=1200000000
temp=58.5'C
frequency(45)=1200000000
temp=59.6'C
frequency(45)=1200000000
temp=60.7'C
frequency(45)=1199998000
temp=61.2'C
frequency(45)=1200002000
temp=61.8'C
frequency(45)=600000000
temp=62.8'C
frequency(45)=600000000
...
frequency(45)=600000000
temp=77.4'C
frequency(45)=600000000
temp=77.4'C
frequency(45)=1200000000
temp=78.4'C
frequency(45)=600000000
temp=77.4'C
frequency(45)=600000000
temp=77.4'C
================================================================================
T/V                N    NB     P     Q               Time                 Gflops
--------------------------------------------------------------------------------
WR02R2L2        8000   256     1     1              97.34              3.507e+00
HPL_pdgesv() start time Fri May  6 22:41:49 2016

HPL_pdgesv() end time   Fri May  6 22:43:26 2016

[Rive]

The thing throttles back at about 62 degrees and maxed out at about 78 degrees

I am confused by this...Did you set the threshold lower? or are you on a cpu governor other than 'ondemand'? Because the Pi3 should not start to throttle back until it hits 80C.

This may otherwise indicate a power/ voltage issue that you have yet to resolve. I had an issue where I was running @ 600 MHz, and it was because I was using a 5V 2A power supply. Soon as I replaced it (with 5.25V 2.4A power supply), the Pi3 ran at the ondemand freq of 1200 MHz until it throttled at 80C. An obvious indicator may be a blinking/flashing red LED just prior to the non-heat related throttling event.

When the Pi3 has power/voltage issues, it WILL default to the lowest freq (600 MHz) setting available...which is what your Pi3 appears to be doing with the benchmark.

A power supply issue will most likely also explain why your Gflops are only 3 (i.e., your whole device just slows down, reducing the Arm processor capability)

[gregeric]

Edit: Unless there is some other regulator for the SoC that I am not aware of.

Hi heater - late seeing this & you already twigged it runs at 1.2V + over_voltage.

The "other regulator" for the core voltage was internal to the soc on the 2835, running off the 5V rail.

There's a separate 3V3 & 1V8 reg which supplies Vddio etc.

On the 2836 the internal reg for the core supply was removed/not employed, we got the xenon-flash sensitive external one in lieu, controlled via ic2.

There's a similar external core regulator on the 2837, only not photo-sensitive.

The 5.1V figure seems spurious.