Keep in mind I'm not trying to argue that PS3 is better or worse than the 360 (I develop on both), but rather go through some of the PS3 advantages which KZ2 took advantage of which can be talked about in a public forum (ie backed by stuff which has been publicly disclosed). Clearly had KZ2 been a 360 game, I'd be talking about 360 advantages instead.
30% seems really high, unless you meant how much cpu use it would take. I don't think a gpu would take that long post processing the items you mention. I agree that it's not a bad choice to do it that way on the PS3, but I don't think anyone can make direct comparisons to other machines because it doesn't make sense to do it that way on a unified gpu.
30% for a 60 Hz game might be a good upper bound. I think there was an Insomniac presentation which quoted a high number like this. Cannot remember if it was for early R2 (possibly a 60Hz target then?), might not make given (I think) the shipped R2 is a 30Hz game.
As long as CPU/GPU memory access isn't a problem (ie your GPU can write out to CPU memory, or has unified memory, or the CPU can quickly access GPU memory), and the CPU has a serious vector unit (which the SPUs qualify for) with enough bandwidth, then doing post processing CPU side can make sense if you have free CPU time. Sure wouldn't try this on the 360 however. If you go way back to the old SGIs (O2, etc) which had unified memory, you might be doing some post processing CPU side. Likely in the future as GPUs get more CPUified, everything will stay GPU. However if PS4 ends up with both a DX11 class GPU and a lot of new SPUs, you might very well want post SPU side just to find enough to keep them busy.
... It's true that texture fetch is a bottleneck during post processing, but you can also use this to your advantage because you can go nuts with alu instructions on a unified gpu during post process with no performance hit. That lets you try all manner of stuff! For example, you can combine what are normally separate post processing steps together. If you have two steps that can use the same 16 taps then glop them together and go alu crazy, it won't matter because all that alu time will get absorbed by the unified gpu running at full tilt. Net result is you can get entire post process passes free.
Agreed, lots of ALU free, bandwidth not free.
This is true. However, from what I recall KZ2 still has a forward render pass for particles. It would have massively benefited from EDRAM there. Resolves are irritating but they are very fast, faster than the performance loss you will encounter doing a forward render transparency pass on a non edram setup.
Yeah, EDRAM is awesome for 8bit blending with lots of overdraw. And it is clear with KZ2's low resolution 4x MSAA trick for rendering particles that particle rendering was an issue (as it is for everyone else as well).
I think you are overstating this one. What you are saying is true, reusing shadow buffers on 360 would be a pain. But it's moot because you wouldn't do that anyways. You have to do that on PS3 to get around the heavy memory handicap that machine has, on 360 there's piles of extra ram so just keep 4 separate shadow buffers around, no big deal at all. Plus, there are potential advantages to keeping the shadow buffers around.
Sure advantage of shadow buffers being around are for static shadows, or if you don't update all your dynamic shadow maps every frame. But you still eat the memory cost for that. Seems like with KZ2 that they use dynamic fast changing shadows to their advantage, like the back-projected muzzle point shadows. Likely they are also taking advantage of hardware PCF, and the fact that shadow map generation is rarely fill limited on small shadow maps (as you might have in a shadow cache), and instead fill a common shared larger map for higher quality shadows all around.
The PS3 "heavy memory handicap" is bogus. Knowing from public docs that people can do post SPU side should provide enough hints as to why this is so. Also one would have to be making a lot of assumptions about texture bandwidth, physical texture layout in memory, texture cache, and other important considerations which we won't ever be talking about (for obvious reasons) to validate or disprove any such over-arching comment. Better to keep it as that both platforms have various advantages and disadvantages with respect to texturing and memory which can support a very similar level of quality (which IMO both Gears 2 and Uncharted have shown on their respective platforms).
