This router's ability to flabbergast me has no bounds.
The UART cable came in today. I open the router. Soldered pins to the serial connections. Monitored the terminal. Started a large steam download at 28.2MB/s.
Nothing.
No crashing what so ever.
I can only think of two things that it could possibly be at this point.
I'm leaning towards the first.
Very strange!
Are you able to put the top on loosely with the serial still connected and test again?
The nbg6817 is indeed running considerably hotter than its competition, like e.g. the r7800.
May have been one of the reasons it was discontinued in 2021.
Unlikely, all other ipq806x devices have been discontinued in favour of ipq807x; nbg6817 and r7800 were pretty much the last holdouts.
I found this post supporting the thermal theory. His earlier post hypothesized that changes between OpenWRT v18 to v19 might have put more load on the CPU than had been present before since he didn't run into the issue on v18.
Don't know if this would work on a NBG6817...
It does, thermal sensors are present and supported.
And I just enabled graphing. The router sits around 62c idle, and gets around 72c in areas on those high download loads. That's with the top of the enclosure off.
I can't check mine right now, but it's quite a bit hotter (closed case).
That situation was quite different in 2017, when I bought mine - and it's still a good choice. For me, it has weathered an unexpected over 5-times speed increase on WAN.
This was a GREAT video!
With the enclosure on, even somewhat loosely with the serial cables peeking from the side, my temperatures rose to a minimum of 74c idle and peaked at around 85c where I started to see the 5gz radio crash again. The kernel didn't produce any console errors or anything at this point, but the ap just crashed and had to be reset from the web interface to start working again.
So with that I think we've finally narrowed down what is happening with this router...
85c is really toasty. I don't let my PC processor get that hot lol.
I've ordered some heatsinks off of Amazon and am going to do some testing to see if this helps mitigate the heat, both with the enclosure on and off. It appears that the 2.4ghz radio is on the far left of the device and under a ventilated area of the enclosure, but the 5ghz radio is slightly off center and under some structural elements that allow heat to gather, rather than dissipate.
I'd like to put a fan above it or something but I'm not familiar with where I could draw power for one.
The easiest places to locate, would be the barrel plug (12V) or the USV ports (5V).
Looking at this teardown photo, the r7800 clearly did more for thermal mitigation than the NBG6817: https://openwrt.org/_detail/media/netgear/netgear-r7800-front-pcb.jpg?id=toh%3Anetgear%3Ar7800
I imagine those thermal pads connected to a metal chassis or something?
Would you believe the idle temperature and under load temperature dropped to about 38-40c lol.
I don't plan on keeping it this way, but a fan is definitely needed. I was hoping that opening up the side with the added heat sinks would be enough but the heat still got into the 80's without airflow.
My plan is 3D print a frame for the opening with raspberry pi sized fan mounts to press fit against a plastic mesh to cover the opening.
Update on my router overheating dilemma... I don't know how I didn't notice this before, but there is paper between the chip and the heatsink...
They made the oldest mistake in the book 
At first I thought this was a thermal pad but its not... Its the wax paper protecting the thermal pad or adhesive... This very well may have been the culprit and I really went all out already unfortunately.
This same paper appears to be in the teardown photos as well!!!
Could you get it off in a reasonable way (given that the heatsink is basically soldered in place)?
