heruamon said: He also says this:
"Even here at NBGI, the Soul Calibur Team gave up on it for Soul Calibur IV. In order to accomplish this effect, we had to write a lot of code optimised to make full use of the PS3’s SPU. Ultimately, we implement the same motion blur on the 360 so this will be a great challenge"
so, they pushed the ps3 to the limit? |
I believe he is suggesting that, for the SC IV project, the team was unable to achieve the same blur effect on both consoles, so they removed the feature from both.
He's stating that his team is giving it another run, and that they have accomplished what they set out to do on the PS3, but not yet on the 360 -- but they are trying.
Motion blur is probably best done, for their game, with multiple character instances rendered to multiple render targets with assorted grades of alpha and blurring effects on each image, and then the images are composited for the final scene. The multiple character instancing and additional animation processing would be something faster (not to be confused with "easy") to do on the PS3. I would wager that any single-image shading effects would be just as easy, if not easier to do, on the 360... but thats as long as you had heaps of vertex data to begin with, and compositing multiple images isn't something you're shooting for. Compositing multiple images might be again be easier on the PS3, due to some framebuffer/rendertarget swapping issues of the 360's GPU.
Think of it this way. They are pro
bably going for a really good version of motion blur -- they are probably rendering, and animating, each of the two characters fighting onscreen a bunch of times. If there are a lot of characters, with varying degrees of alpha and shader effects, animated at 60 frames-per-second (each character instance would be in a slightly different pose, so its as if there are, lets say 12 chararacters onscreen at once, instead of 2), and they have large skeletal bone counts, as well as high vertex skinned meshes... you can believe the Cell would gobble that kind of processing up, relative to the Xenon.
The Xenos (I'm talking GPU now, not the Xenon CPU), however, could be much speedier at rendering the individual instances of the character -- that is, once all those skeletons are done posing for the frame, and all those skeletal structures have been skinned. As is, they probably (I'm really guessing here, so take this with a grain of salt) are having to use more of the Xenos pipelines for vertex processing than they would like to for skinning, and the 3 vector coprocessors of the Xenon CPU are probably taxed trying to "keep up" with 6 SPEs doing what the SPEs are best at -- hence their problem. The RSX isn't held up by this issue, since most of the vertex processing has already been accomplished by the SPEs, and its free to just run off with the pixel shading/rendering.
This is just my beer talking though. I'm a procrastinator by nature, so I probably don't know anything. The last bit he stated isn't very good english, so really its kind open to interpretation, isn't it? ;)