You should sneakily edit the quoted post and your own post and then accuse HIM of a sneaky edit! (Thats what I would do anyway)

That was a pretty cool read.

I didn't understand about 80% of it, but it was pretty cool.

i am reading it.

SPU aren't co processor, but general purpose processors.

loads of DMA calls, and it looks like they program  in C. c++

@ Jo21

IMO the most important tips to be learned from all these Insomniac, Naught Dog, Guerilla Games, etc articles.

- The SPUs are very efficient and powerful while they can perform any function you want from an ordinary CPU (they are general purpose processors) to do and are powerful/flexible or competent for traditional GPU functions as well. Ideally in the end all game engine code running on the PPU now, should in the future be moved onto the Cell's SPUs.

- Keep your code portable. The SPUs require some special care to code for, but when the code is properly ported it runs far more efficient both on the SPU and PPU (and any other CPU architecture for the matter). Portability can be important for debugging purposes and of course for cross-platform developers.

- Don't optimise your code too soon. The SPUs are extremely powerful and efficient, even a game like Killzone 2 has tons of processor cycles to spare so they can add even more SPU systems without having to optimise existing code. Optimisation can be done at a later point if so required. There are lots of oppertunities for optimisation, dramatically increasing the SPU performance.

The EDGE toolset sychonizes the SPUs automaticly... that's awesome, but they still are not general purpose CPUs... nothing he can try to say will change that. They are called SPUs for a reason. Though I will agree that they are not co-processors. This will be very useful for PS3 only games that is for sure.

SPU stands for Synergistic Processing Unit.

Synergism just means: The interaction of elements that when combined produce a total effect that is greater than the sum of the individual elements, contributions, etc.

They are general purpose processors and are more feature rich than processors of the past which were also considered being general purpose processors. They are extremely powerful for calculations, which is a common trade for a high spec processor.

Apparently they're using pipelining (ppu->spus->gpu) to use the additional cores without doing so much work... the obvious consequence is an increase of response time over the traditional method (1 additional frame of lag or 33 milliseconds if running at 30 fps).

It really looks like multicore the lazy way, but I guess it's a good tradeoff.. just nothing impressive or scalable in terms of multicore programming.

Thanks! These presentations are always a good read. Can't wait for this game!

They use MFC?

You mean the memory flow controller?

How can they not use it? Isn't it the only way to transfer data to/from the SPUs?