It's been a long time since I look at machine language but, if I understand correctly, the difference is the capacity of the operation. Instead of performing a

It's been a long time since I look at machine language but, if I understand correctly, the difference is the capacity of the operation. Instead of performing an addition operation on 1 or 2 bytes at a time the operation can be performed on 64 bytes at a time. Given that 'deep learning' requires a number of parallel simple operations expanding the volume of parallel operations in a single CPU instruction boost capacity. Seriously, just supposition until someone with better understanding chimes in.