shams said:
Hus said:
HA.
Ubi talking PS3 dev, how would they know anything they suck at it.
Ai is mathematics, it pretty well known thats what Cell excels at. You just got to program it the right way. Those fools should give Insomniacs a call. Next Splinter Cell is a 360 exclusive, theres no reason why it can be compared to Killzone or MGS4 |
Going to ignore(?) the rest of these posts, but I don't agree with this. There is mathematics, and there is mathematics. Operations such as square root (or 1/square root) are expensive, and heavily used in physics. The AI that I have done in the past only uses "basic" maths (weighted averages and the like), and a lot of comparisons, branching, option evaluation and so on. Small amounts of operations across a large and random data set (rather than intense calculations, possibly on a smaller data set - such as triangle rendering, or skeletal animation/blending). I don't know exactly what sort of Ai they are talking about - possibly some form of weighted response calculation (i.e. to determine their next decision, maybe path finding, etc..) - but I would definitely prefer to write it on a 360 than a PS3. As Entroper says - it can be developed (efficiently) for the CELL, but it requires more thought, effort and time. Generally less time worrying about the AI itself, and more about the architecture of the hardware. (in a way, this defines the approach that MS takes to everything - abstract the hardware away, and worry about the problem at hand. In theory you get lower potential (max) performance, but its a lot easier to write - and takes a lot less time. So you can be left with more time to optimise the algorithms themselves, rather than the architecture code). |
huh that's true, Microsoft is a software company, and "abstracting the hardware away" is mostly what they do. never thought of it that way, but it's kind of obvious now that you spell it out.
sony of course is an electronics/chip maker (though exiting the chip business), their emphasis won't be on software engineering principles.
and yeah, i just want to chime in that distributed programming is not easy to code... requires lots of thought and very hard to maintain. the research group i work in has this parallel simulation code that was developed in the 90s that was tested on running on over 1000 processors... but after the grad student graduated the parallel version of the code has fallen into disrepair (as you add new functions to the code, the parallel stuff need to reflect those updates, but it's very time consuming and error prone). nowadays, we turn to outside programs, developed by groups dedicated to speed and efficiency and with good maintenance support, to run similar simulations.
more fundamentally, not all algorithms can not made parallel. i don't know what Ubisoft has in mind but when in doubt, more than likely it won't be worth the develoment money to do this parallel AI coding. i'm sure programmers themselves would welcome a challenge, but managers won't like the increase in cost.
