Heat Sinks, Do YOU need them.

[emba]

im guessing the chip doesnt contain a temp monitor as we have yet to see any actual temp figures, im not sure if ill have time this weekend as i have a few things on but im going to try and get something setup to monitor the temps soon.

for those that have used heatsinks which are you using? im looking at the akasa gpu ram sinks but if anyone has any other suggestions im open.

[carlosfm]

Why do people feel the need to overclock/volt ?

Is it a macho thing?

Oohh, great question.
It must be a macho thing, yes...
I speak for myself, I always search for some extra speed, but with rock solid stability.
It takes time experimenting, but then you settle to the "ideal" values and no more messing around.
In the 80's I've changed the Intel 8088 CPU at 4.77 Mhz (!) in my IBM PC XT for a NEC V20 and replaced the oscillator on the motherboard for a faster one.
I've always overclocked, these days it's so much easier, so why not?

BTW, capacitors are usually rated at 85ºC or 105ºC, they can typically handle some 2000 to 4000 hours at these extreme temperatures.
A hot device can't be reliable, in the long term.
The Pi, even when not overclocked/overvolted, is already uncomfortably hot.
Notice that the CPU gets to 100% utilization very easily and stays there for a while, even when doing basic things.
The ethernet chip is always hotter than the CPU/GPU/Ram chips (even if you overclock/overvolt them), and the regulator after the micro USB power plug gets hot too.
Does the Pi need cooling?
For me, it's mandatory. But that's me.

[permaband]

my pi has been up 10 days now operating as a webserver, even under full load it only gets 'warm' and much of that is I suspect the acrylic case's poor conductivity. I see no need for heatsinks or active cooling of any kind tbh. certainly not in stock trim.

[ejsolutions]

@carlosfm: +1

[bredman]

It is 8 pm here in Susanville and it is 76 F inside with the cooler is at full and 88 F outside when I got into my truck at 6 pm the inside temperature according to the truck was 143 F inside the vehicle and 94 F outside so If I install my Pi in the truck it certainly would not hurt to at least heat-sink it.

Just because it is uncomfortable for a human does not mean that it is uncomfortable for your RPi.

The CPU/GPU chip on the RPi is rated for air temperature up to 85 degrees Celsius with no air movement. At this temperature, the silicon will be running at 120 degrees Celsius. This is hot enough to boil water (but not much) and is still within the standard operating range for this chip.

In your truck at 143F (62C), the RPi would be well within its operating limits. You should only worry if your truck will be hotter than 185F (85C).

Note: To push the CPU/GPU to 35 degrees above its environment, you would have to run both the CPU and GPU at 100% constantly.

[jnd]

Do you have a link to the data sheet for these temp limits? 100c seems high to me 150c is just crazy. I'm not saying they don't but I have never seen a semiconductor operate correctly at those temps.

These values usually mean the junction or silicon temperature which can be significantly higher than the case or package temp. I'm not sure how is it with these highly integrated devices these days but simple linear regulator can be used to 125 or more C, depending on the manufacturer. Most of them have some kind of thermal protection which kicks in around these temperatures. Some old power transistors in TO-3 metal package are even specified working up to 200C. Their high power ratings are of course just theoretic and assume that the package can be kept perfectly cool at 25C or even less. Also, you have no headroom for other components or inputs tolerances.

I know that desktop CPUs have relatively low temp limits and I suspect this is due to high current leakages and thermal runaway concerns because of the specific process technology, basically cheaper high performance trade off for limited home use.

[16-32]

Some of the GeForce 7800 graphics cards go over 100c and I think that was within their intended operating temperatures.

Do you have a link to the data sheet for these temp limits? 100c seems high to me 150c is just crazy. I'm not saying they don't but I have never seen a semiconductor operate correctly at those temps.

These values usually mean the junction or silicon temperature which can be significantly higher than the case or package temp. I'm not sure how is it with these highly integrated devices these days but simple linear regulator can be used to 125 or more C, depending on the manufacturer. Most of them have some kind of thermal protection which kicks in around these temperatures. Some old power transistors in TO-3 metal package are even specified working up to 200C. Their high power ratings are of course just theoretic and assume that the package can be kept perfectly cool at 25C or even less. Also, you have no headroom for other components or inputs tolerances.

I know that desktop CPUs have relatively low temp limits and I suspect this is due to high current leakages and thermal runaway concerns because of the specific process technology, basically cheaper high performance trade off for limited home use.

[blc]

Why do people feel the need to overclock/volt ?

Is it a macho thing?

For me - with standard desktop hardware - it's about getting more bang for your buck. If your chip can hit higher speeds reliably then why the heck not? I mean, the chances are that you're going to replace it long before chip's expected lifespan elapses, even if you account for any lifespan reduction caused by overclocking. I'm not really into the "extreme overclocking" stuff like liquid nitrogen/helium cooling though (frankly that's a criminal waste of helium in my book, but that's a topic for a different discussion); that is sheer electronic phallus enlargement, if you ask me.

I built my current PC about three and a half years ago and it runs a Core2Duo E5200 rated at 2.5GHz. First thing I did (after installing an OS) was to overclock it; after a few stability tests with Prime95, I settled on an overclock of 3.75GHz. If I had better (i.e. 1066MHz) RAM, I could probably push that well over 4GHz.

An extra 1.25GHz of performance for free. That's why I overclock

As @AndrewS has stated some of those who believe in using heat sinks are old overclockers that believe every chip has to have a heat sink on it. And he may be right. It cost me $2.25 to put them on mine. (actually it cost me $20 for 10 big ones and 10 small ones plus shipping)

I haven't overclocked my Pi yet but it does feel a bit wrong running it without a heatsink! I know full well that it can handle being "naked", but it still makes me nervous!

[Mortimer]

So you have a heat sink glued to the back of your phone do you?

[blc]

So you have a heat sink glued to the back of your phone do you?

Of course not

But in many tablet/phone/SFF devices, metal EMF shielding can also act as a passive cooler - assuming it's direct contact with any processing hardware. A tablet I had last year used this idea, as well as the Atom netbook I used to have. Even if it doesn't have some form of thermal compound, it's going to conduct at least some heat away.

Besides it's a bit different when there's a naked processor staring at me from the desk, tempting - nay, daring me to touch it and see how hot it is...

[jamesh]

Feel free to touch away. Then you'll see how hot it doesn't get.

Cannot comment on the ethernet chip though although that should'nt need a heatsink either.

[adlambert]

Regardless of what my Pis are doing, I have never felt anything more than tepid warm on any of the components.

[Mortimer]

But in many tablet/phone/SFF devices, metal EMF shielding can also act as a passive cooler - assuming it's direct contact with any processing hardware. A tablet I had last year used this idea, as well as the Atom netbook I used to have. Even if it doesn't have some form of thermal compound, it's going to conduct at least some heat away.

I would be very surprised if EMF shielding is in direct contact with the chips. You wouldn't want any inductive coupling in the shield, as then there would be little point in it being there. In all the mobile devices I have managed to take a look inside of, there has been a gap, albeit very narrow, between the shield and the components inside it. So if anything the EMF shield is going to restrict even further the what little airflow there would have been inside what is in most cases going to be a nice, thermally, insulating plastic case. So again, if you show no such concern about having a heat sink on the chip in your phone, why do so for the SoC on the Raspberry Pi? It doesn't make sense to me.

At the moment, I think the only concern for heat should be directed at the USB/Ethernet chip, which if some posts are to be believed, may be supplying power to the whole board when it shouldn't be, resulting in them getting hotter than usual.

[bredman]

At the moment, I think the only concern for heat should be directed at the USB/Ethernet chip, which if some posts are to be believed, may be supplying power to the whole board when it shouldn't be, resulting in them getting hotter than usual.

The USB/Ethernet chip normally runs at around 60 deg C, which would be considered hot to the touch. I saw somewhere a measurement by IR thermometer which showed it running up to 62 or 63.

But this is well within the design limits of this chip. And the temperature of the USB/Ethernet chip will not change if you overclock/overvolt the CPU.

[blc]

So again, if you show no such concern about having a heat sink on the chip in your phone, why do so for the SoC on the Raspberry Pi? It doesn't make sense to me.

But I haven't got a heatsink on my Pi, and I don't plan to. I haven't even seen the need to overclock it yet; I don't do the same sort of tasks with my Pi that I do with my PC (where I really do need the extra processing grunt). I might test some overclocking soon, when I start testing some emulators, but even then I've no plans to slap a heatsink on it...

All I said was that it just feels wrong to have a computer processor sitting there without a heatsink. I know full well that it's designed to be run like that - if it needed a heatsink it would have shipped with one - but it just goes against everything I've learned over the last 15 or so years of working/tinkering with computer hardware. Especially when I've fried processors in the past because I didn't mount the heatsink properly.

Can't really explain it any more than that - it just feels wrong!

[Tooms]

At the moment, I think the only concern for heat should be directed at the USB/Ethernet chip, which if some posts are to be believed, may be supplying power to the whole board when it shouldn't be, resulting in them getting hotter than usual.

The USB/Ethernet chip normally runs at around 60 deg C, which would be considered hot to the touch. I saw somewhere a measurement by IR thermometer which showed it running up to 62 or 63.

But this is well within the design limits of this chip. And the temperature of the USB/Ethernet chip will not change if you overclock/overvolt the CPU.

The USB/Ethernet chip is get warm because of an design error where the LAN9512 chip is feeding the CPU/GPU/RAM with the 1.8V rail.

I have posted images of that here http://www.element14.com/community/message/57800#57800


Tooms

[Lob0426]

So you have a heat sink glued to the back of your phone do you?

The point you are missing is that there is no room to put an "effective" heat sink on my phone. An iPhone by the way. Or I might just heat sink it.

Seriously though equating the issue of heat sinks on Pi's to placing them on cell phones is like comparing apples and oranges. One, my cell phone will shutdown and give me a warning that it is too hot to use. Th RasPi has a temperature sensor but it is not available to us through software. It may be required to stay on if it is used in an automotive application. So heat sinks in that use just make sense. In Britain maybe not, but here in California yes. The SoC really does not get that hot, but the LAN9512 does. And even though it runs within spec on your desktop it will not stay in spec inside a car compartment that is over 140F (60C). Besides how can you go and buy a nice new heat sink for the LAN chip and leave the star performer (SoC) naked like that? It is unseemly!

So again Heat Sinks are not required for the normal operation, I.E. desktop programming, of the Raspberry Pi. But some of our Pi's just are not being used normally! I have one that is being "Back-Powered through a USB hub. It has no PSU attached to it. It is powered by the hub through a single cable. Working just fine after switching out fuses. It has heat sinks.
http://www.raspberrypi.org/phpBB3/viewt ... 50#p144250
I really do not see that as a "Normal use" of a RaspBerry Pi. My other one is running a Lapdock, but it has been dressed up with heat sinks anyway, I had them why not use them. I had to RMA #2 it had a SD card problem. It would not write to any partition without some corruption. When I get the return it will probably end up in a project box with a wireless router. It will have heat sinks on it too. The box will be ran closed with no ventilation. So I will hedge my bet a little towards stability. Might not need them but it will get them anyway.

It seems people are either for them or totally against them. Why no middle ground here?

My Panda Board got a set too and it really does not need them either supposedly!

By the way, if you have not picked up the newest kernel yet make sure you do. It almost corrected the bad RasPi's write problems, just not quite. things are continuing to evolve around the Raspberry Pi world!

[alexeames]

It seems people are either for them or totally against them. Why no middle ground here?

I think some of us are in the middle. But if I'm going to put jewellery on my Pi it's got to be made of precious metal.

[FX4]

Do you have a link to the data sheet for these temp limits? 100c seems high to me 150c is just crazy. I'm not saying they don't but I have never seen a semiconductor operate correctly at those temps.

These values usually mean the junction or silicon temperature which can be significantly higher than the case or package temp. I'm not sure how is it with these highly integrated devices these days but simple linear regulator can be used to 125 or more C, depending on the manufacturer. Most of them have some kind of thermal protection which kicks in around these temperatures. Some old power transistors in TO-3 metal package are even specified working up to 200C. Their high power ratings are of course just theoretic and assume that the package can be kept perfectly cool at 25C or even less. Also, you have no headroom for other components or inputs tolerances.

I know that desktop CPUs have relatively low temp limits and I suspect this is due to high current leakages and thermal runaway concerns because of the specific process technology, basically cheaper high performance trade off for limited home use.

Yes you are correct, power transistors handle high heat. I misstated myself. I was referring specifically to asics, in particular microprocessors and package temperature, not junction temperature. As I stated before it's been about fifteen years since I worked in the semiconductor industry but our device would fail if we exceeded junction temperatures of right around 125C. For reasons I can't remember now we specified the upper operating limit at 90C.

[emileigh]

Enjoyed reading this discussion - I have overclocked and I do have heatsinks. I have been worried about heat and yes, you can find out the temp:

vcgencmd measure_temp

Mine were at 60c before I overclocked and 62 overclocked which I felt was too high. (I panic if my PCs CPUs - two Intel i7 3770K's - go over 45). With the small heat sinks it is running at 50.3 under load overclocked. I too was a hardware engineer, and I firmly believe heat kills. Getting the heat away from the chips is good practice IMHO.

Cheers,
Emi
(PS - I also got a better power supply (1.8A - it's from Blackberry and it's 5.95$US at our favourite e-tailer. Clean power is good power.)

[Lob0426]

I still believe the heat sinks will extend the life of your RasPi. Especially where there temperatures are hot, like in my high desert region.

Are they a necessity? No.
Do they hurt anything? No.

Just make sure they stay attached. Metal moving around on a circuit board is not good!

[rpdom]

Enjoyed reading this discussion - I have overclocked and I do have heatsinks. I have been worried about heat and yes, you can find out the temp:

vcgencmd measure_temp

Mine were at 60c before I overclocked and 62 overclocked which I felt was too high. (I panic if my PCs CPUs - two Intel i7 3770K's - go over 45). With the small heat sinks it is running at 50.3 under load overclocked. I too was a hardware engineer, and I firmly believe heat kills. Getting the heat away from the chips is good practice IMHO.

And the new prize for resurrecting an old thread goes to...

Heatsinks are still not necessary. 60°C or so is nothing to worry about. It may shorten the life of the chips by a few months, even years, but will you still be using your Pi 25-30 years from now?

My Intel P4M laptop tends to run at around 62°C. It is at least 8 years old and going strong. The fan (a bit noisy now) cuts in at about 65°C. The highest I've seen it run at is 75°C. Some of my AMD Athlon servers run at 80°C+ and I've never had a CPU fail on any of them.

"Getting the heat away from the chips is good practice IMHO"
Yes, for the pointless heatsink manufacturers/sellers

[jamesh]

The BCM2835 will run fine up to 85degsC, at which point it will throttle itself back to reduce heat. I'm not sure anyone has ever reached 85 except in exceptional circumstances (In a metal box in a desert perhaps!)

So, no need for heatsinks. Although I do this this on my device...



Just for giggles.

[Heater]

That's no good. You haven't got a fan on it.

[mikerr]

Why do people feel the need to overclock/volt ?

Is it a macho thing?

No heatsinks for me - the lan chip gets mighty hot though
- most of my pi are at stock speed,
except for one:
my main pi which is at 1000MHz and also just as importantly (for speed) runs usb root.

Not bothered about speed on pi's that are single purpose/ssh e.g. remote camera or file server - but it makes a large noticeable difference when using it interactively in X , or compiling