As noted in the previous Alpha Topaz Video AI 5.3.0.1 Alpha (Hyperion Alpha #3 + RXL Alpha #1):
What they likely mean by “RXL is a larger variant of Rhea” is that it has more parameters/weights/layers. This means the model has more numbers that need to be stored in VRAM to process the video frames and thus it consumes more memoy.
Memory usage can be decreased by using smaller tile sizes. But in the last Alpha it seems the system that picks which tile size to use based on avaliable VRAM was incorrect and was picking tile sizes that were too big and consuming too much VRAM. This may still be a issue in this Alpha.
