It should be clear that each perspective goes significantly on the memory. A simplified calculation looks like this:
Front perspective = 1 GB
+ Diffuse perspective = 1 GB
+ Rear perspective = 1 GB
------
= 3 GB
======
For each register with tremulant, the amount for one register doubles.
If memory is scarce, you can proceed as follows:
- omit one or more perspectives (do not load)
- load a relatively dry perspective (front) in 16-bit
- load several pedal registers in 16-bit only
- load only one loop
- do not load tremulant registers (you can use the virtual tremulant)
- load all registers in 16-bit
You may have to do without rear, although I have found that even a slightly mixed rear signal increases the sound volume and opens up the room.
If the above measures are not sufficient, then in case of hardship you can use only the front channels and shorten the reverb tails and supplement them with IR. However, I have made the experience that if you use a foreign IR, then you remove the "soul" of the organ, because the organ was intonated to the room and only sounds optimal with the original acoustics. This does not apply to relatively dry samplesets, where you can best take advantage of an IR.
This is a pain in the ass trial and error, but HW unfortunately doesn't offer memory calculation in advance.
With my 64 GB I sometimes have to apply some of the points too, but that's only the case with very large sets.