To test the setup I looked for the cheapest second-hand CPU the motherboard could support - found a Ryzen 5 3500X. From the benchmarks I actually don't see a big point in upgrading to a Ryzen 5000, unless the difference will be huge or I specifically want an APU (like in the original design). My GPU is around 700 grams (with backplate), so if I don't mind the extra weight I could use one PSU and have the choice of CPUs without integrated graphics.
An apu is nice to have to do this hybrid gpu thing that some laptops do to save power on idle. Not sure if it's worth giving up a good cpu though.
Also apus seem like the only real way to get a multi gpu setup, given the size and power constraints.
I'm thinking about 60-90cm from the ATX PSU to HDPLEX, so not a huge extension - just something to have the PSU a bit further away from my lap.
With a 6mm2 cable you should get around 0.1V drop and 4W loss at 40A or 0.05V drop and 1W loss at 20A. The hdplex or the boost converter shouldn't care much about the drop. A dc-atx plug shouldn't mind it too much either as long as the psu isn't outputting something very low like 11.5V.
ATX spec for 12V says ±5%, I think. So 11.4V-12.6V? Not sure how things react on either end of this range.
I'll have to think of a sensible battery solution to keep everything alive at least for a graceful shutdown if not for a sensible amount of work time. Power draw at the plug was ridiculously off, a good "worse-case scenario" - GPU alone displayed 275W under load, but Dell DA-2 is 220W max and my GPU is 175W TDP. Draw at the plug was max 155W when booting, 105W idle, 410W under load.
105W at idle seems really wrong. What do you use to measure it?
Maybe it's displaying apparent power and not true power?
Power draw and battery run time for comparison:
Load: 55W -- 40min (stopped load when one cell reached 3.2V)
Idle: 30W -- 100min (system shutdown when bms cut out)
(12W of it is just for the display alone because I don't know how to dim it)
The load was Prime95 running on an a8-9600 APU.
The power draw was measured at the battery using a clamp meter. The battery has old 2200mAh cells in 6S1P configuration.
The gpu (rx480, 150W tdp) draws
110W under load (or 210W, I can't read my handwriting. Either way it seems wrong)
20W idle, output connected
12W idle, output connected, nothing displayed
0-3W idle, output disconnected. [measured with clamp meter only]
I don't expect max load on the battery pack so 4S2P should be enough - just enough juice to unplug and plug in somewhere else, or save gam... work and gracefully shutdown. The problem is stepping it down to 12V at that current.
Something like this claims to support 40A output at 12V.
https://www.ebay.co.uk/itm/DC-DC-Sp...tep-Down-Buck-24V-auf-12V-40A-AH/154279726680
It's hard to tell if it will actually handle it though since none of the components are visible. Though with a 18-35V input range it is not going to work with a 4S setup anyways. It should work for 6S, 7S and 8S.
4S2P is already more cells (and 2x BMS to charge) than 7S1P. Any 1P solution feels neater to not balance each parallel pack separately. 8S1P would also be an option for more reliable high load off of battery.
There is no need for additional balancing in a 2P setup if the cells are wired as series of parallel pairs. Most laptop batteries do it like this.
EDIT:
Looking at any DIY battery packs and balancing, especially with multiple in parallel, I'm tempted more and more to buy a Zippy or Turingy pack, cut it open and rearrange to be flat so it fits in the bottom box - even 7S wouldn't be a problem then
Sounds good. How many cells would fit in total this way? Hopefully the battery pack doesn't have the cells connected with spot welded metal strips.