For those of you who own Pro or even use 5.0.4, I noticed that with the latest 9950X, aiming to get the full bandwidth of the PCIE, and cheating in every way possible. (splitter, adapter, riser etc…)
Threadripper 2990WX (best in TR4)
Asus ROG Extrem Zenith Alpha (best in X399)
→ 64 GB/s
All lanes are fully configurable, you can adjust them as much as you want up to Gen3 x4 for each GPU, which is enough (an RTX 4090 goes up to 2 GB/s bandwidth for me)
Ryzen 9 9950X (best in AM5)
Asus ROG Crosshair Hero (best in X670E, maybe not ^^)
→ 64 GB/s (out of the available 108 GB/s)
Due to the poor PCIE lanes splitting capabilities, you can’t even take advantage of the full bandwidth of PCIE Gen 5+4 
The only advantage is PCIE Gen 5 Nvme.
Also, the memory latency impact seems low → Went from 6000 MT/s 48-48-48-96 to 6000 MT/s 30-36-36-90: Windows is clearly faster but the framerate in TVAI is the same.
