GA107 should be significantly larger vs NAVI 24 (107mm2). I estimate GA107 closer to 170-200mm2 range.
What would be nice to know is not so much actual die sizes (or transistor count which might be more accurate between TSMC 7nm and Samsung 8nm) but die costs: we all asume that Samsung gave Nvidia a sweetheart deal for 8nm but by how much?
Still the full GA106 is 12 billion transistor according to TPU (GA107 just has that down as "Unknown", ditto die size), while Navi 24 is 5.5 billion. Now RTX 3050 is cutdown and should eventually really be on GA107, but shader unit wise TPU has:
GA107 = 3,072 shaders
GA106 = 3,840 shaders
Obvioulsy there are other cuts, but that implies GA107 is around 80% of GA106.
Navi 24 = 1,024 shaders
Navi 23 = 2,048 shaders.
So by that Navi 24 is pretty much half a Navi 23.
So yes, like you said, there really should be a chip in between. Navi 24 really looks like it would compete with a GA108 if Nvidia ever make that.
Pretty sure AMD didn't plan for a chip between N23 and N24 so no wonder they ran the 6500 XT at nearly 3GHz (and a crazy 100W). If they cared, the should release a 8GB Navi 24 card but they'd first have to source some 4GB GDDR6 chips (if they even exist), or reduce the price.