lol. The "Velocity Engine" (a vector processing unit) isn't actually even true 128-bit. Rather, it is a SIMD (single instruction multiple data) unit that can process up to 128-bits of information each cycle but the largest chunk of data it can process is 32-bits. SIMD means that given 4, 32-bit pieces of data (128-bits in all) that have to have the exact same instruction applied to them, rather than having each one go through the processor in its own turn, the SIMD unit can just apply the same rule to each one. Say that you have co-ordinates for the four corners of a box: {(3,4), (3,8), (5,4), (5,8)}. Now, you move this box along the x-axis and shift it over 4. Your resulting box would now be {(7,4), (7,8), (9,4), (9,8)}. Now, in a traditional non-SIMD fashon, the processor would first have to add (3,4) + (4,0) to get (7,4) and then have to procede to the next data piece (3,8) so on and so forth until it finished it. Now, the "Velocity Engine" just applies the same "+ (4,0)" to evey piece at the same time in one processor cycle rather than having to do each one in turn. This has just saved you 3 cycles of the processor.
The "Velocity Engine", though, can do more than just apply it to four 32-bit operations. It could also scale to do eight 16-bit, sixteen 8-bit, and I think it scales even smaller than that. So, if you were having to do the same thing to sixteen 8-bit pieces of data, by using the "Velocity Engine", you would be saving the processor an extra 15 processor cycles in which it could be doing something else.
The only problem with this is that usually the only time you have SIMD instructions are in graphical manipulations b/c when you shift a box or something like that, you are shifting all the points of that object and all of the points connecting the corners of the object. But we all know that graphical manipulation isn't the only thing the computer is used for, technically, the "Velocity Engine" isn't worth much most of the time. It is better than nothing, though. Intel has the MMX, SSE, and SSE2 instruction sets for its SIMD operations which deal more with streaming media than graphical manipulation.
Note: the example I gave will not be entirely accurate to all of you who know about vector processing, but I am merely trying to put it into easier terms to understand. lol. Also, I may not be on the boards all that often, but I hope the info I provide is useful and worth the breaks in attendance
