Here is a picture of them without the flash so you can see the LEDs on the Pis. There is also a red LED at the end of each Blade Duo indicating they have power, although you can't see them in this photo as the red power plate covers them.
These 8 Pi3 model B+ have been running for a week or so now and haven't given any invalid results so far. I put that down to the better power supply than my USB chargers.
This is the power supply. These are normally used to run 12v LED lights in your ceiling (sometime called down lights) but this one is a 24 volt 10 amp output to make sure there is enough juice for all 20 Pis.
And this is the way they stand when one puts the rack parts around it. The top and bottom bits of the rack screw into the metal standoffs that you can see.
There is another pair of power plates and their 5 Duo's which stand next to this. However before I can do that I need to order more Pis.
3 comments:
Hi, nice stuff. I have a BR40 based project - see http//www.enigma.fi/oh2bnf
Such great hardware and fun to work with. I run mine via PXE which has not always been without its hiccups, but does offer certain benefits over to SD's.
Best regards,
Mikko
Which model Pi do you have in your rack? I believe tmy Pi3 model B+ are thermal throttling as they take about 1 hour longer than my other Pis when doing the same work.
I have plain vanilla Pi 3's except for 3 - 4 of those new Pi 3 + versions. In regards to the heat management, I have noticed that it is indeed well reasoned to have fan assisted cooling; the enclosure has provisions to mount 4 small fans, which are likely sufficient. I am however using a little bit different - and temporary - cooling setup, consisting of 6 USB-powered fans pushing air directly from the front of the enclosure (the blue cover is currently not in place).
In fact past week or so I have been specifically pushing the envelope to see the temperature behavior, running the 160 cores at full load 7/24. I did not (yet) however take any measurements in regards to the job execution time.
As time soon permits I will setup proper MPI layer and begin to prepare the system for industry standard benchmarkings. I should also look into the "weights" of the Pi 3+ in comparison to the standard Pi 3's, at least in the scheduler (SLURM), but for all practical purposes, I seem to have a slightly asymmetric cluster.
Decision whether to use a standard Pi 3 or a Pi 3 + was entirely based on what the stores had on their shelves at the time I was buying them in batches. Now that you mention throttling, I may have actually have observed some anomalies while watching the cluster do its work. Could be related to (throttling). It would be interesting and fairly easy to set up a measurement routine which would reveal the "heatmap" within the enclosure.
Sorry it took me a while to respond, I had received an email about your comment in the Gmail which I rarely read.
Post a Comment