The original Denise (8362) does not have this register, so
         whatever value is left over on the bus from the last cycle will
         be there.  ECS Denise (8373) returns hex (fc) in the lower 8
         bits.  Current AA Lisa revision return hex (f8).  The upper 8
         bits of this register are loaded from the serial mouse bus, and
         are reserved for future hardware implentation.

The 8 low-order bits are encoded as follows:
+------+----------------------------------------------------------+ | BIT# | Description | +------+----------------------------------------------------------+ | 7-4 | Lisa/Denise/ECS Denise Revision level(decrement to | | | bump revision level, hex F represents 0th rev. level). | | 3 | When low indicates AAA feature set (???) | | 2 | When low indicates AA feature set (LISA) | | 1 | When low indicates ECS feature set (LISA or ECS DENISE) | | 0 | Maintain as a 1 for future generation | +------+----------------------------------------------------------+
There are some caveats in using this register in AGA testing. The original 8362 Denise didn't have this register at all! That is, it returns random¹ values when it's read. So there is no liable way to check AGA through this register! Better way to do it is to check AGNUS/ALICE revision. The best way is to check it from graphicsbase. YOU CAN NOT rely on multichecking of this register!! Results depend on system load, DMA activity and such unreliable things...
Note¹: Non-existing hardware register returns value that happens to be present in hardware data bus at the moment. When DMA is running there seems to be somewhat random values. This can be prevented: only thing you have to do is to kill all DMA for a moment and then get the value. For all the 8362 chips I have tested the result has been $ffff (currently 1 chip :-). I cannot guarantine this, though. Delta/PitPlane
Interesting links: