Small fans are notorious for failling quickly. You'd be better off using an oversized fan at moderate rpm, which would consume about the same power as a smaller fan at higher rpm. The bigger fan would probably push more air, although the pressure and directionality of the smaller one is better. But cooling the few Ws of a Pi is easy with just any airflow.
MrS
Agree. I had looked into buying a 2" or 3" clear pvc pipe. Mounting the Pi in the pipe and attaching a 80mm fan to one end and a filter at the other end. This would definitely create a "wind tunnel" and assure adequate cooling. I am so impressed with the guys/gals who have the laser cutters for punching out Pi cases, but that is a bit pricey. After all the whole reason to have a Pi is to have an expensive computer. Hmm, maybe a 3d printer could make such a "tunnel". Now you might think this is childish but a fan that is illuminated by leds is a real plus. At night and somewhat during daylight hours a quick glance can let you know that it is turning. Stationary blades are not good. All of my big crunch cases came with illuminated fans but I used to turn the illumination off. Not anymore!!
I posted this response to "MrS" about 8 mos ago and alluded to using PVC to fashion a "tunnel" to cool a Pi. Recently I purchased a 3D printer and after much effort believe I have reached a point of relative comfort with this printer. I decided that my first project would be to build a PiTunnel/Kooler. Here are some pics of that first effort.
My goal is to build a 7 cell (2 over 3 over 2) Hive where the Pi in the center cell monitors temps in the other cells and controls their respective fans.
Currently my Pi(s) are in use so I am awaiting the arrival of another Pi. When it arrives and I put all this together I will make the CAD files (.stl) for this build available as a download on my site for anyone interested.
Where would all the cables go? I guess in the "floorspace" between the bottom of the tunnel and the thing that the PI is screwed to? In that case maybe it would be cool if this flat "floor" piece would extend over the full length of the tunnel, with some cut-outs to thread the cables thru. Not sure if I made myself clear, but it would be a miniature version of the "raised floor" concept, sort of :-)
Also a version where you can put a few (4 perhaps?) PI Zeros into a single tunnel would be really cool. Just in case you are looking for another challenging 3D printing project. The only difference would be different "floor" blade to push the PI Zeros in, I guess.
On another note: in my spare time I was experimenting how long a job for the (currently in beta) Gravitational Wave search would take on a Raspberry PI 2. Those jobs will run ca 10 hrs on a reasonably modern Intel CPU core, on a Raspi Pi2 core it will take .... about 10 days :-) . I tried to see what kind of speedup you get if you use a multi-threaded FFT, but so far, the relative short length of the Fourier transforms (ca 2^17) and the threading overhead appear to make this not very efficient. Using the GPU based FFT might be worth trying as well.
Well, 10 days will probably make the deadline we will be using in the actual (after testing) run. It's more like a fun thing, being able to claim that a Raspberry Pi will contribute to the search for Gravitational Waves from spinning neutron stars.
Once the testing is done we will merge back our testing branch into the public repository where the GW search code can be found and anyone interested can try to speed things up for the Raspberry PI2 or Parallella board, just for fun.
Two GW tasks with X64 ended in about 48 hours on my Opteron 1210 at 1.8 GHz running SuSE Leap 42.1, the latest SuSE release. It is an old but reliable machine, running 24/7 since January 2008.
Tullio
Where would all the cables go? I guess in the "floorspace" between the bottom of the tunnel and the thing that the PI is screwed to? In that case maybe it would be cool if this flat "floor" piece would extend over the full length of the tunnel, with some cut-outs to thread the cables thru. Not sure if I made myself clear, but it would be a miniature version of the "raised floor" concept, sort of :-)
The tunnel/cylinder length is 95mm. The floor (Pi Plate :>) ) is 90mm so you can "wrap" cables under the floor as you note. I will be using a Wifi dongle for networking so I am a bit concerned with how it will work inside. It might need to be placed "outside" (another access port). When my Pi arrives I will no doubt realize that there is a need for more enhancements/changes/access ports.
Quote:
Also a version where you can put a few (4 perhaps?) PI Zeros into a single tunnel would be really cool. Just in case you are looking for another challenging 3D printing project. The only difference would be different "floor" blade to push the PI Zeros in, I guess.
I have not looked at the dimensions of the Pi Zero. But it might just require a new "plate" as you say so that you could mount 5 (???) side-by-side by 2 rows in a vertical orientation as long as the length on the PiZero cards don't exceed 90mm.
Going forward I would not print the cylinder part. It is a 6 hour print and if you have a power failure you get to reset the clock and start again. I was lucky on Thursday, but lost a part Friday due to a power surge. It was a small part so not a big deal. Also these printers should not be left unattended. No you don't have to sit next to it but you need to be "around" just in case you have a thermal runaway. I am aware of only one person who had to rebuild. The extruder nozzle runs at 223C and the bed at 60C for the material I am extruding.
My first Pi cruncher was an early Pi. I had configured it so that /boot was on the SD card and everything else on a USB hard drive. I was a reliable cruncher. My question: In your Pi configurations do you put everything on the SD card? Any problems with corruption due to constant downloads/uploads, reads/writes? Using a USB drive would require a larger "cell" footprint or a cell just for USB drive wired into the Pi. Another access port. The possibilities are endless.
I have not looked at the dimensions of the Pi Zero. But it might just require a new "plate" as you say so that you could mount 5 (???) side-by-side by 2 rows in a vertical orientation as long as the length on the PiZero cards don't exceed 90mm.
Its 65 x 30 mm. Thinking about dimensions...would you be able to fit two PIs in that tube , one at the "floor" and one at the "ceiling"? And likewise perpaps 8 to 10 PI Zeros? With all the effort going into printing this stuff and an abundance of airflow, it seems reasonable to try to increase the package density as much as possible. A package of 10 PI Zeros would actually justify a need for a more serious cooling :-)
Quote:
My first Pi cruncher was an early Pi. I had configured it so that /boot was on the SD card and everything else on a USB hard drive. I was a reliable cruncher. My question: In your Pi configurations do you put everything on the SD card? Any problems with corruption due to constant downloads/uploads, reads/writes? Using a USB drive would require a larger "cell" footprint or a cell just for USB drive wired into the Pi. Another access port. The possibilities are endless.
All my BOINC crunching PIs use SD cards. And yes, the cards will eventually fail, but I haven't bothered to find a better solution.
I have not looked at the dimensions of the Pi Zero. But it might just require a new "plate" as you say so that you could mount 5 (???) side-by-side by 2 rows in a vertical orientation as long as the length on the PiZero cards don't exceed 90mm.
Its 65 x 30 mm. Thinking about dimensions...would you be able to fit two PIs in that tube , one at the "floor" and one at the "ceiling"?
Two PIs would fit without problem by stacking one Pi on top of the other using nylon standoffs. There is insufficient horizontal space to accommodate a Pi on the bottom of the plate if I understood you correctly. With two PIs stacked there would not be sufficient space remaining for attaching a motor/servo hat. The reason I went with an 8 sided cell rather then a six is that it provided for more of a clean vertical stack.
Quote:
And likewise perhaps 8 to 10 PI Zeros? With all the effort going into printing this stuff and an abundance of airflow, it seems reasonable to try to increase the package density as much as possible. A package of 10 PI Zeros would actually justify a need for a more serious cooling :-)
Quote:
My first Pi cruncher was an early Pi. I had configured it so that /boot was on the SD card and everything else on a USB hard drive. I was a reliable cruncher. My question: In your Pi configurations do you put everything on the SD card? Any problems with corruption due to constant downloads/uploads, reads/writes? Using a USB drive would require a larger "cell" footprint or a cell just for USB drive wired into the Pi. Another access port. The possibilities are endless.
All my BOINC crunching PIs use SD cards. And yes, the cards will eventually fail, but I haven't bothered to find a better solution.
On a crunch box with all the upload/download/read/write activity I would much rather go with the /boot partition on the sd card and the rest of linux on an SSD. I have a write up on how to do this here if interested. I was looking at SSD drives as a choice for a hard drive. They make a USB to SSD cable which could plug into a Pi2. The Pi Plate could be modified to provide taps for both the Pi2 and the SSD drive. The plate could also be reversible such that it provides PiZero slots on one side for vertical mounting of the PiZeros. With Pi Zeros you have to make certain assumptions in design. Namely that everyone will use cables sized to fit a Pi Zero. With their small USB connector footprints gender benders would be a problem unless you make the "tunnel" huge. And this adds to the print cost and print time.
Two PIs would fit without problem by stacking one Pi on top of the other using nylon standoffs. There is insufficient horizontal space to accommodate a Pi on the bottom of the plate if I understood you correctly. With two PIs stacked there would not be sufficient space remaining for attaching a motor/servo hat.
I was thinking about mounting the Pis not stacked but opposite to each other, using two plates and therefore two "cable floors". But yes, no hats will fit in then. I was thinking about number crunching only.
RE: RE: Small fans are
)
I posted this response to "MrS" about 8 mos ago and alluded to using PVC to fashion a "tunnel" to cool a Pi. Recently I purchased a 3D printer and after much effort believe I have reached a point of relative comfort with this printer. I decided that my first project would be to build a PiTunnel/Kooler. Here are some pics of that first effort.
No. I did not print the fan
If interested you can view the complete effort here along with some 3D printer videos
My goal is to build a 7 cell (2 over 3 over 2) Hive where the Pi in the center cell monitors temps in the other cells and controls their respective fans.
Currently my Pi(s) are in use so I am awaiting the arrival of another Pi. When it arrives and I put all this together I will make the CAD files (.stl) for this build available as a download on my site for anyone interested.
Neat! Where would all the
)
Neat!
Where would all the cables go? I guess in the "floorspace" between the bottom of the tunnel and the thing that the PI is screwed to? In that case maybe it would be cool if this flat "floor" piece would extend over the full length of the tunnel, with some cut-outs to thread the cables thru. Not sure if I made myself clear, but it would be a miniature version of the "raised floor" concept, sort of :-)
Also a version where you can put a few (4 perhaps?) PI Zeros into a single tunnel would be really cool. Just in case you are looking for another challenging 3D printing project. The only difference would be different "floor" blade to push the PI Zeros in, I guess.
On another note: in my spare
)
On another note: in my spare time I was experimenting how long a job for the (currently in beta) Gravitational Wave search would take on a Raspberry PI 2. Those jobs will run ca 10 hrs on a reasonably modern Intel CPU core, on a Raspi Pi2 core it will take .... about 10 days :-) . I tried to see what kind of speedup you get if you use a multi-threaded FFT, but so far, the relative short length of the Fourier transforms (ca 2^17) and the threading overhead appear to make this not very efficient. Using the GPU based FFT might be worth trying as well.
Well, 10 days will probably make the deadline we will be using in the actual (after testing) run. It's more like a fun thing, being able to claim that a Raspberry Pi will contribute to the search for Gravitational Waves from spinning neutron stars.
Once the testing is done we will merge back our testing branch into the public repository where the GW search code can be found and anyone interested can try to speed things up for the Raspberry PI2 or Parallella board, just for fun.
Two GW tasks with X64 ended
)
Two GW tasks with X64 ended in about 48 hours on my Opteron 1210 at 1.8 GHz running SuSE Leap 42.1, the latest SuSE release. It is an old but reliable machine, running 24/7 since January 2008.
Tullio
RE: Neat! Thanks.
)
Thanks.
The tunnel/cylinder length is 95mm. The floor (Pi Plate :>) ) is 90mm so you can "wrap" cables under the floor as you note. I will be using a Wifi dongle for networking so I am a bit concerned with how it will work inside. It might need to be placed "outside" (another access port). When my Pi arrives I will no doubt realize that there is a need for more enhancements/changes/access ports.
I have not looked at the dimensions of the Pi Zero. But it might just require a new "plate" as you say so that you could mount 5 (???) side-by-side by 2 rows in a vertical orientation as long as the length on the PiZero cards don't exceed 90mm.
Going forward I would not print the cylinder part. It is a 6 hour print and if you have a power failure you get to reset the clock and start again. I was lucky on Thursday, but lost a part Friday due to a power surge. It was a small part so not a big deal. Also these printers should not be left unattended. No you don't have to sit next to it but you need to be "around" just in case you have a thermal runaway. I am aware of only one person who had to rebuild. The extruder nozzle runs at 223C and the bed at 60C for the material I am extruding.
My first Pi cruncher was an early Pi. I had configured it so that /boot was on the SD card and everything else on a USB hard drive. I was a reliable cruncher. My question: In your Pi configurations do you put everything on the SD card? Any problems with corruption due to constant downloads/uploads, reads/writes? Using a USB drive would require a larger "cell" footprint or a cell just for USB drive wired into the Pi. Another access port. The possibilities are endless.
RE: It is an old but
)
Wow, that IS reliable!
RE: I have not looked at
)
Its 65 x 30 mm. Thinking about dimensions...would you be able to fit two PIs in that tube , one at the "floor" and one at the "ceiling"? And likewise perpaps 8 to 10 PI Zeros? With all the effort going into printing this stuff and an abundance of airflow, it seems reasonable to try to increase the package density as much as possible. A package of 10 PI Zeros would actually justify a need for a more serious cooling :-)
All my BOINC crunching PIs use SD cards. And yes, the cards will eventually fail, but I haven't bothered to find a better solution.
RE: RE: I have not
)
Two PIs would fit without problem by stacking one Pi on top of the other using nylon standoffs. There is insufficient horizontal space to accommodate a Pi on the bottom of the plate if I understood you correctly. With two PIs stacked there would not be sufficient space remaining for attaching a motor/servo hat. The reason I went with an 8 sided cell rather then a six is that it provided for more of a clean vertical stack.
On a crunch box with all the upload/download/read/write activity I would much rather go with the /boot partition on the sd card and the rest of linux on an SSD. I have a write up on how to do this here if interested. I was looking at SSD drives as a choice for a hard drive. They make a USB to SSD cable which could plug into a Pi2. The Pi Plate could be modified to provide taps for both the Pi2 and the SSD drive. The plate could also be reversible such that it provides PiZero slots on one side for vertical mounting of the PiZeros. With Pi Zeros you have to make certain assumptions in design. Namely that everyone will use cables sized to fit a Pi Zero. With their small USB connector footprints gender benders would be a problem unless you make the "tunnel" huge. And this adds to the print cost and print time.
It's a SUN M20 Workstation.
)
It's a SUN M20 Workstation. Its front panel is open, allowing a good airflow. That must be the main reason for its reliability.
Tullio
RE: Two PIs would fit
)
I was thinking about mounting the Pis not stacked but opposite to each other, using two plates and therefore two "cable floors". But yes, no hats will fit in then. I was thinking about number crunching only.