This board is somehow flawed. If it is T-Topology, it would make no difference using B2/A2 or B1/A1 on two DIMM setup. But it does. With two DIMMs XMP 3200 doesn't work stable with Crucial Ballistix Micron E-Die. But with four DIMMs, it does work. And not only 3200 XMP, I managed to make it stable up to 3533 fast timings - I settled down at 3466 fast because 3533 fast needed 1.46V.
In 3200 XMP with 4 DIMMs, motherboard configures ProcODT 60, RttNom 34, RttWr 120 and RttPark 240. 24-20-24-24 for the MemCadBus values.
Those settings are similar to what Ryzen DRAM Calculator suggests when running 4 DIMMs Dual Rank though. Such settings don't go higher that 3266, but if I use ProcODT 48, RttNom 34, RttWr Off and RttPark 48, 24-24-24-24 for the MemCadBus values, I can run up to 3533 with 4 DIMMs stable. A simple change of ProcODT to 53 makes it unstable immediately.
With only 2 DIMMs, setting above values for termination makes it almost stable at 3200, but not 100%.
With Zen 2 CPU you can easily get stable up to 3533 with 2 DIMMs. I think the reason is the termination. Zen 2 runs with a much lower termination for ProcODT. This board seems to work best with Zen/Zen+ with ProcODT of 48, although this is not good enough for just two DIMMs.
So, to conclude, I think there is a problem with this board, the termination of the DIMM slots and the algorithm used for RAM training. It doesn't make sense to be able to run stable at 3533 with 4 DIMMs and failing to do so with 2 DIMMs, even at much lower speeds. On a T-Topology, assuming the IMC can handle the 4 DIMMs, 2 or 4 DIMMs should run at same speed. Also running B1-A1 or B2-A2 should make no difference. On a Daisy-Chain, B2-A2 should be faster than 4 DIMM setup. So this board behaves completely different to what is expected for either T-Topology or Daisy-Chain.