I can “me too” this thread. I agree, the arch thread you reference is the same issue. I first noticed HDMI failing over 2 years ago.
For my part I am on MBP 15" 10,1 2012, gentoo, kernel 4.13.9, and nvidia-drivers 387.22. This is a GT 650M. Connecting monitors directly to the HDMI port results in detection via xrandr and nvidia-settings, but no signal to the connected screen. Connecting thunderbolt->VGA or thunderbolt->DVI does properly drive the same monitor. OSX drives all the same monitors via all adapters. The nouveau driver drives all the same monitors via all adapters, including HDMI, as long as the monitors are connected at boot. This makes me think the nouveau driver is leveraging EDID detected from EFI possibly?
Over the years, researching this problem has been frustrating because of the uniqueness of the symptoms. But now it seems a lot of folks are running into it and are fed up enough to describe it thoroughly. This thread in particular, and the reference arch thread, are some of the best descriptions I’ve seen, so thank you!
I’ve also run across this:
https://wiki.archlinux.org/index.php/NVIDIA/Troubleshooting
wherein the “xrandr BadMatch” section describes “… the proprietary NVIDIA driver clips the pixel clock maximum frequency of HDMI output to 225 MHz or lower.” Not sure if this is significant.
But I also wonder if this driver is simply having trouble getting EDID via the HDMI port?