If you have already 3060 ti, you can depopulate that 2699v3 down to just 1 CPU, but you need 256GB RAM and you get 3 min/plot (if that box is PCi3, faster on PCIe4). Also, my understanding is that MM GPU plotter is happy with just 4-6 cores, so you can get a less power hungry 2600v3 CPU, and still do the same. I have v2 CPUs, and those take DDR3, so 256GB is about $150. I think that you need DDR4, and that will cost you more. I plan to try using 2600L v2 version (low power) to further reduce power consumption (it costs about $10 on eBay). I have not started it yet, so have no data for that. Of course, you have v3, so that will cost more, but it is potentially worth to look into.
Having that setup, I think that just 1 NVMe will be enough as a buffer. Max also has plot write parallelizer that you can use to push those plots from that NVMe to few drives. My understanding is that you will need around 4 drives (not sure if 3 will cut). This means that you would need to check how much your USB controller can push, whether it will be able to keep all those 4 drives at full speed.
Unfortunately, 64 GB RAM (on that 5900x) is not going to let you use GPU plotter (you need at least 128 GB + fast NVMe - not sure if in RAID0 setup). This way you will be able to use the CPU plotter. Max updated his base MM plotter, and it can give you a boost of ~10% or so over the old version.
So, having those two boxes you mentioned on hand, I would try to really depower that v3 box down to bare bones, as it could potentially plot 5-10x faster than any 128 GB based box.
I think that there is no info about chia’s GPU plotter, neither about when nor about what.
Saying “cheap” doesn’t really mean much, as the more plots you need to replot, the more sensitive your setup is to power draw during replotting and less to the initial cost.