The motherboard I finally chose was the ASRock - TRX50 WS AMD sTR5 eATX Motherboard, which has the sTR5 socket I needed for the threadripper and 5 PCIE slots... If you want to run multiple GPU you need a CPU and motherboard that has enough PCIE lanes- I am HOPING that the next generation of Threadripper will use the same socket type so I can upgrade the CPU at least once- boy for my work it is the GPUs that matter most- My last computer had three GPU (NVIDIA 1080), the 4090 alone should outperform all 3 1080 put together by a huge margin...

For AI the VRAM on the GPU is very important- 4090 has 24Gig vram, and the 5090 will have 32gig VRAM with a huge leap forward in VRAM bandwidth

This is also the first computer I have owned where some of the chips were made using the EUVL process I worked on at Sandia National Lab way back in 2001... it took ~20 years for that Chip lithography process to overcome all the technical challenges to be used commercially- a single EUVL machine costs ~400 million dollars...
For my build:
The CPU is ~ $1300
4090 GPU is ~$1900
Motherboard is $700
a 96 gig stick of RAM is ~$550
1600Watt power supply
I went with a lot of SSD as well
Its a large expense, you can get a top end gaming computer for significantly less, I last bought a desktop maybe 8 years ago... I don't do this often ... Currently you can expect that most of the open source AI applications will take a lot of time to install and get working, they are not simply things you quickly download and use- many/most will only run in a LINUX environment, and you download the source code from GITHUB... each one will likely take a lot of learning to use.

I expect that will change rapidly with more user friendly applications coming out soon, but those will likely be costly too ... I hope to use opensource stuff as much as possible which is free but takes a significant time investment.

HOPEFULLY I will get to the point that I could make a high quality video advert for the HZ using these tools and have it sound like its narrated by Morgan Freeman ( https://github.com/WadRex/RVCompact ).... THAT will get more members! :)