I tested a dozen of heat sinks in many dimensions, none worked because of the missing airflow. I even bought ones that have dimensions 15x10cm with fin height of 6cm (because i thought maybe its possible to cool the case more). Passive cooling is not possible.
Now, i use the sinovoip fan which already has a heatsink - also, i use heatsinks on the sfp port. On both (cpu + sfp) i have a tiny fan, one noctua and one from sinovoip. But even with those AND the heat sinks, it did not work. Now i replaced the top of the case and all of sudden, it works like a charm. It works that well that i believe, replacing the complete oem fan of sinovoip with a bigger heat sink could be enough. The holes on the side (oem case) are just bad because one has to suck in fresh air which leads to a bad mounting position when using regular fans. If you want to use the holes on the side and NOT create new holes at top/bottom, you should probably go with something like on laptops. Otherwise, i highly doubt it works or it must create such big amount of air flow, that its just… stupid because of noise.
edit: just noticed you updated your pictures. that should work as you go exactly the way i described.
But i am still not sure if that would be sufficient to cool sfp+ as this goes crazy hot. I burned my finger on it, so i would say it was about 80°C or even more - with heatsinks.
Continue your effort. If you find anything usable, publish it. I am out of this discussion now.
You are right. I forgot to mention that there are mounts for a 120 mm fan, a 2.5 inch disk (with cable minimum length 30 cm), 8x8 led matrix on WS2812 to display any events (I haven’t thought of any yet) and the router board itself on a removable panel. I will post the design soon and give a link here.
In terms of air circulation, it is much better than the original case… Could you provide 2 holes in the back (where the RJ45 ports are) for antenna pigtails? I have 10 antennas (6x WLan and 4x LTE).
Could you also publish the files as DXF? I have the option of laser cutting. Only they need the templates as DXF…
Thank you for your work, @Stonty. Here is another idea you might be interested it. There is a problem with the stacked PCB design of the R4, which causes strong RF interference from the main board to the Wi-Fi module (check this thread). One solution would be to move the NIC away from the main board using two extension adapters like this one. I haven’t used these before, but I’m willing to check if they work.
Looking at the mezzanine add-on for the antennas in your design, there should be enough room to do so, possibly flipping the NIC upside down to keep the heatsinks ventilated. Beside the practical reasons, making the wireless part self-contained like that just seems cool to me.
I think my design could easily be adapted to your idea (just add a rectangular hole to pass the FPC cable through - this can be done with boolean operations in most software). To me, I guess the main problem with your idea is PCIe signal integrity - you may need to use a retimer or even a redriver in the extension module.
Flexible PCIe risers for PC work fine unless the length is crazy. I’ll keep your warning in mind, though.
I might add that in your design the main PCB is upside down when the mezzanine is used. That is not ideal for SFP modules and RJ-45 jacks, as the locking mechanisms end up facing down (SIM trays too). Connecting the wireless NIC with flex cables could change that too. Instead of both PCBs sitting in the middle of the air flow they could form the floor and ceiling of a wind tunnel, possibly reducing the overall height.
Actually, it’s upside down on purpose. I first tried to put it upright, but I found that the heat sink put too much pressure on the NIC card and it bent (I used a zip tie to fix the heat sink).
To be honest, I think your idea is quite great. I also considered a similar idea when I was designing at the beginning, but finally gave up because of its complexity. In addition, regarding your mention of “Flexible PCIe risers for PC work fine unless the length is crazy”, in fact, although most PCIe devices have a high tolerance for PCIe signal integrity (I have even seen applications that transmit PCIe signals through USB D+/D-), I still recommend using best practices to avoid any problems. Especially from the past history, I don’t think that BPI’s PCIe design redundancy may be sufficient.
Then it remains to be seen if the extenders work. If I end up exhausting all shielding options and the RF noise is still bad, that would be the only thing left to try.
