I'm subscribed to 2 ISPs, 1 Gbps each. A simple test showed that Raspberry Pi 3b+ can do 60 Mbps, whose CPU scored 186 in PassMark (ran that one myself). If I compare that to Raspberry Pi 5's score, the latest Pi should just about saturate a 1 Gbps connection.
I tried N100 before. The coil whine is not acceptable. NanoPi R6S is twice the price for 1/3 more performance, so it's a no-go too. Is there any other option comparable to a Pi 5?
Since you've listed the most performant ARM based SBCs: R6S and Pi 5, and ruled out the Intel N100, the only other I can think of is the new BPI-R4. It has a MediaTek Filogic 880 (4x A73s) at 1.8GHz, might be able to hit your performance needs.
=) I knew someone's gonna bring this up. That's my main router actually. Pi 5 has 4 A76s running at 2.4GHz, which means R4 should score about 2000 in PassMark. That's not fast enough unfortunately. Guess there isn't a better option than Pi 5.
Mm. Should take it up with the vendor/change vendors haha=P
Also helps to narrow down where the coil whine is coming from?
There's always the "I can do power electronics at home" solution =P
some options
change the inductors...
crack the ferrite to change the resonance....
try using glue on the inductors....
change/select a converter that has a switching frequency out of the audible spectrum....
"if it were me": put it in a case / DIY my own case?
Or go for an itx server motherboard and custom case?
We haven't defined power consumption, form factor, and other noise characteristics. edit: so budget is 50-80USD?
Perhaps using an audio spectrum analyser of the existing switching noise would be a good start.
There's also the "put it in another room" solution?
"use a laptop"?
"used market"?
an i3 8100/9100 or i3-10100t gets you the cpu passmark you're after for example?
I thought I was being replied to but not. But here's more stuff about the hacking electronics options
Which part? the power electronics part?
If you're asking me if I've done hacks other than mechanical damping on a multi phase converter for a CPU the answer is no.....
All of those options except for perhaps if the bios has options for the converter would be hard and void warranty.
The glue method only really helps if it's vibration that you can dampen. An easy way to check is to put your finger on the coils (i.e. inductors) and see if the noise goes away. This should be safe as it won't affect the magnetic flux. Inductors can get hot though.... Coils can be rated for >100 degrees c.
Cracking the ferrite if it's a resonance issue will change the resonance. But it will also change the magnetic flux / inductor's electrical characteristics.
Similarly with changing the inductors, would have to see if it's a mechanical defect in the inductor, or all the inductors. Changing the inductor also changes the electrical characteristics of the converter as a whole.
But if you're asking whether I've done any of those hacks myself the answer is no. I just put things in a box, put it in another room etc =P.
The answer is I've built low frequency switchers and they obviously had whine because they were in the audio bands. Plus used switching power converters that go in pulse mode at idle. Anything more than that, just buy a pre-made converter off of parts supplier of your choice =P
Firmware doesn't exactly fix coil whine unless you somehow changed switching frequency and did spread spectrum and whatever. Sometimes it just whines. It's also hearing / user dependent sadly =(
Sometimes it's also when under no/low load you will get the whining as it goes into very low duty cycle mode to save on switching losses....
I don't really care for coil whine as usually my fans are louder.... It's usually switching noise in general not the inductor whine itself that gets me.
Changed the seller, but not the vendor. 8505 faired better in the noise department, but still not acceptable. I didn't poke around the board 'cause I intended to return it, but the noise changed with network traffic. It's especially obvious when doing a single thread speedtest, so I guess it might've been from the ethernet phy (transformer?) and not the CPU.
No, not going down the rabbit hole of making a case. It'll likely cost too much in the end since it's a moving target. Yeah, I can justify about that much. I did record a clip. The spectrogram showed the most obvious noise from an 8505 was ~5.5KHz when doing a single thread speedtest.
It's gonna be in my bedroom, so inaudible is required if it's going to run 24/7, which brings the power budget: 10 watts or less (for low fan speed). A used lappy and such would be power-hungry, big, and not so quiet.
mm. I reckon box within box and some foam would be fine but yes. Design rabbit hole.
Hmm. I guess that's the design then....
mm. I reckon a used tiny PC will be under 20W if you get an older one. But 10W and under is definitely challenging.
IDK about resonances in the ethernet transformer.
Single thread speed test? As in a CPU benchmark? OR as in a network speed test?
Yeah anyway would require more characterisation and opening it up =(
.