It's a good point about the 8700G APU outperforming the RX 6400. The tradeoff between DGPU and APU is hard to get right.
An APU-focused build obviously simplifies just about everything - cost, space, power and thermal requirements...so it depends on how valuable the advantages of a DGPU are to you: not just the obvious performance, but the increased number of options, and the fact that it's easier to upgrade separately.
In thin-ITX land, there's another problem, not in theory, but in practice. There are not many thin-ITX AMD boards, like the X300TM-ITX or X600TM-ITX, and while they are impressively cheap (like
way cheaper than an IMB-1240-WV for example), they lack a couple of important features - no PCI-E slot, which doesn't matter if you're committed to an APU build - but more importantly no eDP header, only LVDS, which I can say is honestly kind of horrible for anything, including games. This would be (and has been) my main challenge with an APU-focused build; how to integrate a worthwhile display, without native eDP. So you might look at eDP control boards, or some other wackiness like Socket Science did with a portable monitor. Looking at mini-ITX, you could also try that ASRock side panel kit mentioned by hrh_ginsterbusch just now, with the one AMD motherboard it supports.
It depends on your space constraints, but I certainly plan on making DGPU battery power delivery feasible for Devotion. The original OpenUPS looks like part of a larger and more generalised solution (where you would have to provide the battery cells), whereas the OpenUPS2 seems to be a smaller, but more specific integration with its own three 18650 cells.
The OpenUPS2 might be enough alone for a low-power thin-ITX system, since you could connect it directly to the internal 4-pin power, though this is assuming you could work around the classic charge-switching problem I mentioned earlier. You could also use it for a low-power mini-ITX system, but yeah you'd need something like a PicoPSU as usual. The charge-switching problem probably doesn't exist in that case, because I guess you just plug your power brick into the OpenUPS2 input, and let it handle the output to the system...could be quite neat. On reflection, I could also do that for a thin-ITX system, and the main reason I didn't is because of the user-experience mess it creates to have a board with a DC jack plugged up because you're not supposed to use it, as well as my own separate DC jack which I would have to put there.
Edit: I also need to say this somewhere, so why not here?...Motherboards draw a variable amount of
power even while switched off, and this includes your normal ATX boards, as well as thin-ITX boards. For the PH12CMI, I measured this as ~1.5W from the wall. For the IMB-1222-WV, it was ~5.5W, which is really substantial, assuming my watt meter wasn't lying. This obviously isn't ideal in a general sense, but especially for a laptop scenario where you might be transporting your system, unconnected to the power brick, and having the battery sapped over time. I believe the MNT Reform
had the same problem, and was able to patch most of it out, because of course they designed the boards. I planned to have a battery isolation switch somewhere reasonably accessible, partly to circumvent this. Not the best user experience having to think about that though, I know. There's probably a more sophisticated workaround to build into a circuit.
still...
