I understand the spec sheet says 12.5 watts per device, and I understand that what I am about to propose may never work.
Our use case requires serval of the fins to be powered by daisy-chaining active or passive PoE.
Now some of you might read this and think daisy chaining active PoE? Who would do that? But ignore that for now.
This post here says that the draw rate on the Pi3B+ under load is 690 mA at 5.2 V so 3.588 Watts.
Even @sqweelygigposted once that he attempted to boot the fin on just 2.7 watts and experienced a brownout quote:
when the Pi started doing anything serious
This implied to me that he got some use out of it at even that very low wattage.
Overall my un-tested conclusion is that fin 1.1 will only draw less than 5 watts peak per device via it’s PoE connection and the real load will be somewhere around 3 watts.
This assumes our use case:
No HDMI
No ePCI router
A low power USB wifi adapter
Is this a terrible idea? Will my devices get deployed and then later be browned out by a software update (that somehow increases the power requirements)?
I will be testing this over the next few weeks and keep this thread updated.
The 12.5W spec on the datasheet is the minimum required to be able to run all the components present on the balenaFin at full capability.
Your application seems to require only a subset of the features and most probably the power requirement will be less. We run some tests with the onboard wifi module and a very basic application running in the compute module and the power consumption was around 4W.
Also keep in mind that extra power will be loss in the PoE HAT when converting the ~48V of PoE to 5V.
What you try to achieve might be possible, it all comes down to what the final implementation is and what, if any, other components are used.
Regarding software updates, it is a possibility that if the devices are updated with a more demanding application, the limited power provided prevents them to boot properly.
Hi…generally a psu performs it’s most efficiently at @50-70% of rated power. So a 500w draw pc on a 700w psu is usually fine. This does not take into consideration psu quality. What you don’t want is a mediocre 550w psu running a 500w pc. It just means excessive wear on the psu and more heat than is healthy over extended periods at almost full load. You also don’t want to run a 1600w monster psu with a 300w draw pc, the load is so low that at idle or barely above, the monster efficiency goes out the window. Normal usage should see no less than 20% and max usage no more than 70% loads to maintain a healthy psu for its expected lifetime.
Hi all,
I will be testing soon and I will keep this post updated with my findings.
We don’t have a choice but to go forward with this for right now. If the tests fail then we will have to reconsider our entire use case. Cable runs can cost 5-20 times the device cost so a home run is not an option.
@ntzovanis I can confirm that 4W reading, infact even accounting for DC to DC loss in the active PoE modules it is still mostly less than 4Ws. I have never seen it peak more than 4.3Ws.
@AraldoJanes
This is interesting, and while this is very useful for me personally (I am about to build my second PC) I am not sure it applies here. PSUs for a PC go AC to DC which as I understand is harder than DC to DC. I don’t have a lot of choices in PoE hats anyways but I will keep what you’re saying in mind when I look at them.
PoE in its many forms is all DC normally 24V to 48V, though ultra PoE can go up to 90. This tends to be low amps, less than 1 is common. On a standard, type B passive PoE connection 30 watts from the source and 25.5 at 100meters for the device(s).
I hope to get 4-5 fins on one line. I expect to be able to get 3-4. 7 would be my goal but I will need to look into ultra for that.
Anyone, please feel free criticize the hell out of my plan. It might save us some money.
What modules are you planning to use to receive and pass on the PoE? I think I personally would try to run a kind of ‘PoE bus’ cable run so at each point I needed a Fin I could cut the cable and introduce some kind of PoE PSU/passthrough device that gives me a 5V power supply. I’d envisage doing this with a device separate from the Fin as it would seem to make the networking aspect easier. I’m assuming you’re using the cable run for communication as well as power here though, if not then maybe some kind of passthrough HAT that does the power without communication would make more sense as you wouldn’t need to introduce any Ethernet switch component into the mix.
Additionally I wondered if you had decided upon active or passive PoE yet? Active of course needing a little more work to implement.
@chrisys
This a lot of fun pushing things in ways they were never meant to be pushed.
Right now we are hard set on passive. Active would be nice to avoid the need of having two wifi interfaces per devices (one onboard for internet, one in monitoring mode usb over pcie). But you would need one of these for each device.
That is both power and cost prohibitive.
However, you are spot on with what we are going to be testing. We are trying to split passive with just coper splitters and use an external DC volt regulator to step it down to 5V and power the fin via the barrel connector.
A passthough Hat would be great but there is nothing on the market for that right now and we don’t want to design it at this moment. Easier to design something external so we can use it with more than just the fin.
Here is what we will be testing for round 1, but in the end we want to get this custom made.
WS-POE-5v10w 24 Volt Input Injector
uxcell 16cm 3.5x1.35mm Female to 5.5x2.1mm Male DC Power Extension Cable Connector
Thank for the input! This seems a little crazy because no one has done it yet. (to the best I can tell)
@tacLog sounds fun indeed - keep us posted with progress.
The only thing I noted from your post is that if you’re using the barrel connector or phoenix connector you do need to provide >6V power as it’s going through the onboard regulator. If you power the Fin from the pins on the HAT connector however, that’s directly connected to the 5V bus and so 5V will work just fine.
Looks like I will have to figure out a new way or strip the wires down.
This might work. I think we will have to make something custom though.
Thanks for pointing that out. I will have to figure this out.
Say I was using the hat. Do the 4 pins are exposed to hats on the side connect to type A as discussed in this diagram? I suspect strongly that they do. But if had my installer crimp the ethernet cable completely wrong then I could route type B injectors (the kind that flood the market) to type A. I think the hats I am using would take that and step it down to 5v for the HAT connector. This could then be used to power the device through the normal enough looking ethernet port.
Part of the struggle here is remote power cycling (remote management that the client controls) and image, in general, is a requirement of many of our clients. So the more custom (or custom looking we get) the harder this will be to sell.
I will keep this post updated and I might become the first to do something! This is not a good thing.
@tacLog if you take a look at the schematics for the Fin v1.1 here, it’s the very last page in the document. You can see exactly how the 4 header pins are connected. I was thinking that you could simply have a connector that plugs into the GPIO header on the Fin to provide power, you wouldn’t have to design/build an entire HAT for it.
Thank you, these are very useful. I had looked at them before but didn’t realize they contained exactly what I needed to know.
It appears that the 4 pin header pins support both modes A and B. This makes me wonder why the PoE hat is not working with passive PoE (type b, haven’t managed to test type a). Unfortunately, the schematics for that are not available online so I may never figure that out.
Our main solution is still a go however. We got the Passive PoE 5 volt splitter today and just need the barrel jack converter. The plan right now is to use the barrel jack to phoenix connector but instead of plugging into the port bend some pins out of the way and plug it straight into pins 4 and 6 on the HAT connector. I know this is a gruesome hacky way to proceed but I want to see if it works.
On the longer term the company, PoE texas also makes a version that steps the voltage down to 24 v or even one that does 19v. That would allow me to use the fins’ onboard regulators.
My goal is to see how many fins we can get powered with a single 30-watt run that uses only one mode( so only 4 copper lines out of 8) then look into doubling up a single cat 6 cable.
Just some thoughts as I am planning this out.
Generally speaking how sensitive are the fins to voltage noise, because there is sure to be some?
What is the difference between using the 5 v bus directly vs going through the onboard stuff?
Also, what would happen if the voltage spiked to say 25 volts via the onboard connectors?
The Fin itself runs mostly from 3V3, so if you’re supplying 5V via the HAT pins that’s getting stepped down to 3V3, 2V8, 1V8, 1V5 for the various use cases on the board. There are some other uses for the 5V bus though; notably USB stuff and the HDMI interface, so you would want to avoid a spike to +25V ideally. One of the cool features of the Fin is the socketed fuses! F4 lies between the +5V HAT connector pins and the on-board +5V bus, so you do have some protection there but it may not be enough to prevent that size of spike causing damage.
However if you were to use the barrel jack or phoenix connector and supply the minimum 6V, it’s going to be able to handle 25V spikes no problem as it’s rated for up to 30V. Perhaps this would be the best option if you’re not confident of the quality of your power?
Caption:
A fin v1.1 powered via its 5v connection on the HAT connector. (The POE hat used is completely not needed) The little black box is the linked device above. It converts the 24 v into 5v out. I am looking to replace with this one that goes 24-48v into 12v out.
I am a bit amazed that it works like this. I hooked it up to an inline POE power draw reader and I have only seen up to 4.4 watts and it seems to rest at around 3.3 watts. This is great news because I am budgeting about 5 watts per device with my goal of 6 devices on one wire. (full writeup on that plan once I test something)
@chrisys
It is interesting that the voltage will be stepped down so many times in my final plan: 48v to 12v to the onboard regulators on the fin which takes it down to whatever is needed. I think this is unavoidable though.
I plan on using the barrel jack and have options for what voltage to provide it. The documentation lists both connectors (barrel and Phoenix) as being 24v max, is that for sustained loads?
Interestingly enough, the fin is also running fin with 5v in to the barrel connection. I don’t know what the risk considerations are for providing less volts to the on-board power regulator. Do you think it would cause harm in the long term?
This isn’t that important as I plan on switching to 12v in to the barrel jack.
I had tested the Fins on my bench with 5V into the barrel jack the other day, and found that it did at least function. However I did talk to @ntzovanis about it, who explained that the dropout of the regulator increases with current. I measured (although this is with a very cheap multimeter!) that when I feed the barrel jack with 5.04V, I see around 4.52V at the HAT connector, this is with the Fin sitting idle (as a balenaCloud device). If I start to make it do some work that 4.52V is going to start dropping further and causing problems.
Your solution of feeding 12V to the barrel jack seems like the best way forward, but you are good to use anything in the 6-24V range on an ongoing basis.
@chrisys
This is very interesting. Would the other voltages that the fin uses be affected or would it just be the 5v bus, which mostly affects USB and HTMI right?
I think we will be working though the on-board adapter though.
I will be documenting and diagramming the solution I have come up with so far, but we are being limited by other design constraints like remote management mostly at this point.
Thomas, any input voltage (apart from the 5V on the HAT header) will be subject to the voltage drop on the regulator. The difference is that since the drop will never reach 1V, then any input above 6V won’t cause the output to be lower than the 5V required for the internal components.
