Deneidez said:
Well, they can use it with simple tasks like physics and stuff(as those are usually embarrassingly parallel), but not really that much graphically as you know today "power = graphics". And when it comes to advanced AI. Making it work on CELL is well... hell, if you can even make it at all. So imo best way how you could enchance games with CELL would be more lively world(moving geometry and calculate for example effect of wind for every object, liquid calculations & so on). |
What you say is so evident ...
Fact is Mike is so occupied dealing with numbers regarding Cell we all know that he forgets real world contingency regarding development process :
1) Not all algorithms are parallelizable.
2) Those are not all the time efficient enough (the gap to mono- core may be not big enough).
3) Legacy of coding : developers already have a very hard time (Look for example J. Carmack's interviews) dealing with multicore CPUs. Imagine the piece of work needed to efficiently work with a paralleled designed one...
4) to build such a CPU, many low level hardware capabilities have been brought back to software. Mike never consider the huge over work that implies. Ints : budget and time to code a game is limited ...
5) You talked about physics on which Cell is particuraly efficient : one of the main point of these calculations is precision. With Cell, you will be able to go further in precision. Will it be noticable in game ? I mean for example is taking account of "exponent 6" argument of the formula really THAT usefull ?







