HI... I am a student and I am currently developing a processor and unknowingly i have designed a completely new architecture. As a consequence i am getting 1440 MIPS at only 45 MHz and scalable... at very low power... I wanted some inputs from you all so as to make it FOR the game ai programmer. Other features: 1)C macro style instruction set, that is same code for all data widths; code and data treated differently in my processor. (i cannot think of the advantages/disadvantages of this as m not experienced in ai programming... please if u can tell me) 2)"globally synchronous locally asynchronous" .... less power consumption and more speed 3)variables rather than registers at assembly level of coding [Edited by - harsh1987 on May 13, 2008 10:35:37 AM]

You have to see what problem types can be solved by parallelization.
The problem usually comes from too fine a breaking up of a problem and having intercommunications of partials states become major overhead.

Im, looking at using clusters to process AI for simulations but data replication required across machines subtracts significantly from the gains.
That is for coarse grained parallelization and the overhead would be much worse for fine grained applications.

Another problem is trying to keep all the parrallel processors busy. Many problems involve narrow narrow channel of serially processed data which leaves most of the rest idle. Pipelining schemes can work for some classes of probems but wont work well for others.

Quote: Original post by harsh1987
<snip> code and data treated differently in my processor.
(i cannot think of the advantages/disadvantages of this as m not experienced in ai programming... please if u can tell me)</snip>

That sounds really interesting, and you should be getting a patent attorney working on it.

The separate program and data memory is called "Harvard architecture". It dates back to the early, early era of computers. See Harvard architecture at Wikipedia.

Quote: Original post by harsh1987
...
1)C macro style instruction set, that is same code for all data widths;
code and data treated differently in my processor.

I think that code be a disadvantage for functional programming.

Quote: Original post by serg3d

I think that code be a disadvantage for functional programming.

how can this be disadvantageous...? that will lead to easy "code cascading" reducing the dependency issue.

Quote: Original post by harsh1987

Quote: Original post by serg3d

I think that code be a disadvantage for functional programming.

how can this be disadvantageous...? that will lead to easy "code cascading" reducing the dependency issue.

Many of the most interesting programming languages used in AI treat code and data exactly the same, so you can assemble bits of code dynamically to produce a program that changes during execution, perhaps based on the data that you use for input. An example is LISP.