Fafalada said:
And what do you think Display lists on PSP are made of? Candy bars?
You have to write the packets all the same, just the command language is a bit different.
This is preposterous, I was assured that Candy Bars were going to be the way to go!
Well, you still have to write less stuff:
PS2 = DMA tags -> VIF tag -> GIF tag
Isn't there support in the PSP API for the DMAC ? That would reduce the amount of code you have to write and data needs only to go from main RAM to GPU (if you do T&L there and all) so you really only needs two tags: one for the DMA controller and one for the GPU interface, right ?
If anyone actually thought that, I think they must have zero experience of any kind with realtime 3d apps. And I maintain that even if the HOS implementation turned out perfect.
I am aware that SCE has great chip designers (at the EE level), but of course the SGI guys at nVIDIA and ATI have paid off for them.
Anyways, I agree with you that 8 MB of RAM compared to 32 MB of RAM is a HUGE jump to fix using HOS.
If they had subdivision surfaces they might have a bit of a better point AFAIK they map better to the geometry people use in games: NURBS + sub-division surfaces would help.
The problem is that, IMHO, 8 MB is still not enough (even taking into account the compression HOS would bring, plus PlayStation 2 developers already compress their geometry a lot [newbies like me still use floats lol] when you take into account textures, program data, etc... especially considering that you do not want to access the UMD drive every cycle streaming stuff in as that will kill the battery life.