Cluster Raspberry Pi power supply from an ATX PSU

Question

I've asked this question within the Raspberry Pi exchange but they stated that posting the question here might be more useful.

I built a Pi cluster with 6 nodes. This runs off a 90 watt astec atx-93 PSU.

Also attached to the PSU is a 12v fan and a 5v ethernet switch.

However it seems the 5v rail of the PSU is not supplying enough current to power everything.

Only 3 Pis boot properly, the others just keep rebooting. And with the Pi's that do boot, attaching any i/o devices to them send them in a spin.

There're four 5v outputs coming from the PSU. 2 Pi's to each output through the GPIO 5v pin with a fuse to protect incase of a power spike.

The 4th output is used to power the fan which is rated for 12v but I have it at a lower speed at 5v.

There is one 12v rail that has a 5v regulator attached. This is powering the ethernet switch.

If I put a multimeter over the Pi's the voltage is 5.12/5.20 v but the current seems to be around 0.9a to 1a max.

So my question is, is there anyway of getting more current to the devices without having to change to the PSU itself, or is it just that buying a more powerful PSU is the only way forward to get what is needed?

Answer 1

According to the schematic posted on Raspberry PI stack exchange, an on-board 1.1A fuse exists between microUSB and GPIO 5v line. This fuse is not supposed to see current drawn from microUSB to connected USB devices. With your method of supplying 5v power to the PI, this fuse DOES PASS USB device current. It is rather difficult to determine in your complex power scheme (where multiple currents can flow in opposing directions) how much current a particular fuse will see. Fuses act on net current flow. Seems like this fuse is running at its limit, if your current measurements are accurate.
Your power supply is adequate, with Neil_UK's caveats about properly loading its other 12v and/or 3.3v rails.
The PI was meant to be powered through the microUSB - it should be powered from this point, else the on-board fuse may not properly fulfill its function.

Answer 2

To me, it seems not your power supply is the problem, but the Pis.

In general, the Pis need 5V and draw as much current as they need. You measured a voltage of 5.2V, which is more than enough. If the power supply were weak (or the power cables were too thin), the voltage at the pi would drop below 5V.

However, the current of 1A seems way to high, for example, this site states less than 500mA for loading the desktop.

I don't know about the other models, but the Model 1 B+ has a 1.1A polyfuse in the 5V line of the micro USB connector. This means, if the Pi draws more than 1A, the fuse blows and needs to cool down to recover. The Pi will reset in this case, and may be caught in a boot-loop. Even if a Pi is just below the current threshold, attaching a device may push it beyond, and the Pi resets.

There's still a chance the power supply sends some voltage spikes, which are not visible on a multimeter. But due to the high current, I don't think that's the case.

As a first test, I would connect the Pis to something different than an ATX supply, and disconnect any devices from the Pi. Next, exchange the SD card to see if there is a software problem.

Answer 3

Looking at the label for your power supply it appears that the +5v and +3.3v supplies come from the same DC/DC convertor block. Notice that the maximum power for both supplies is limited as a pair. In all probability the 3.3v supply is used to provide regulation feedback for both the 3 and 5v outputs.

I'd suggest what is happening is that the 5v supply is actually dropping and you don't see it on the multimeter. There are small differences in the Pi's and the first ones to reset immediately reduce the current drawn so a couple of the units continue to run as the voltage starts to rise again.

I'd suggest you need to apply a permanent load to the 3.3v output. Perhaps start with 10W (1 Ohm resistor). You could also apply a 10-20W load (An incandescent bulb perhaps) to the +5v supply, you might find that that this will prove the +5v is collapsing.