Is everyone here forgetting that 64-bit allows the processor to transfer data twice as fast as 32-bit in bus operations such as DMA? Also, it allows loads and stores of 2 x 32-bit pieces of data or instructions at the same time. Max memory addressability is perhaps the least important difference between 32 and 64-bit processors.
For example for all 32-bit instructions the processor actually receives two instructions at a time allowing it to fill it's instruction caches twice as fast. During memory copy/move operations it is able to copy/move 2 x 32-bit words of data per clock instead of 1. A good percent of program execution actually involves memory copies and moves, especially image/audio processing and codecs.
Another example is say a program task is to go through memory adding 2 32-bit integers in an array. This allows the load of both integers in a single cycle instead of two.
Off base.
The processor gets instructions from cache and will not load two instrcutions into the execution unit.
Cache is filled via the external DDR interface and the size of that DDR bus has nothing to do with cache line size or processor bus width. External DDR bus will not be 64 or or even 32 bits wide.
The ARM64 architecture has the following:
New instruction set, A64
Has 31 general-purpose 64-bit registers.
Has separate dedicated SP and PC.
Instructions are still 32 bits long and mostly the same the 32 bit architecture
THIS IS IMPORTANT -> Has paired loads/stores (in place of LDM/STM).
Most instructions can take 32-bit or 64-bit arguments.
Nothing said about multiple 32 bit arguments.
Multiple arguments would make it a SIMD architecture and for ARM the SIMD stuff is in the DSP extensions. Things like MAC, etc.
Addresses assumed to be 64-bit.
Advanced SIMD (NEON) enhanced
Has 32× 128-bit registers (up from 16), also accessible via VFPv4.
Supports double-precision floating point.
Fully IEEE 754 compliant.
AES encrypt/decrypt and SHA-1/SHA-2 hashing instructions
Also note that the speedup seen in games from A6 to A7 is the newer GPU.
It's Imagination Technologies four core PowerVR G6430.
The older A6 GPU was 266 MHz triple-core PowerVR SGX543MP3
Not sure of the speed but faster and one more core.
So people stop saying the graphics are faster because the A7 is 64 bit.
The graphics are faster because it has a killer GPU.
The Snapdragon uses Adreno GPUs from ATI now AMD, Adreno is an anagram of Radeon.....
Anyway.....
64 bit is not the primary reason it's faster.
The architecture is different with different, more modern GPUs.