* OpenGL ES 3.0 support (things like true HDR rendering, etc. will finally be possible)
* 64-bit support. For most app developers, this will just be a recompile. The tricky part is that any closed-source 3rd party SDKs/libraries/frameworks used will also have to be rebuilt first by their authors.
This was well covered when desktops switched, but the main advantages are that certain operations are faster (e.g. RSA) and the extra address space lets you use mmap() and the page cache for everything, instead of writing your own stupid cache.
Disadvantages are that 64-bit code consumes more memory. There are ways around this.
From the deck: -32bit RISC Instructions -64bit GP registers -64bit operands -64bit pointers
Note that the 32->64 bit register and address size increase is by far not the only difference between 32-bit ARM and 64-bit ARM. Aarchv8 (64-bit arm) is effectively a complete clean redesign of the ISA, designed to be modern, fast, and only maintain some backwards compatibility, discarding many signature features of ARM that no longer make sense. Such as predication of everything and free shifts in every instruction.
64-bit ARM should be significantly faster, at least once the compilers catch up.