The spes being individual and not taking processing power away from the rest of the game is definitely a ps3 advantage. It will in the long run be a bigger ps3 advantage I think.. I guess one has to wonder just how much can be achieved on the 360's 6 threads and how much can a single spu do?

I imagine it can do a lot if it a single spe handles the entire operating system for the console.

Doesn't that leave 5 for the rest of the game? Since there are 6 for available or is another one taken for something?

tsk tsk

1.5 is for the OS
.5 is for sound

5 left for AI, Shaders, Textures and Physics

Isn't ai on the cell handled by the ppe? I was under the impression that the spe had trouble with ai currently or ai just wasn't a good fit for the spe.

That's weird I've never noticed any frame drops in 2k7 videos I've seen then again I don't own the game, but I will.

I have the game and the framerate is VERY solid. Basically 99% of the time everything is fine. I play quarters at 7 minutes, and I notice a framedrop about once per game.

So the game has no problems there at all.

.5 is for sound

tsk tsk

1.5 is for the OS
.5 is for sound

5 left for AI, Shaders, Textures and Physics

I have the game and the framerate is VERY solid. Basically 99% of the time everything is fine. I play quarters at 7 minutes, and I notice a framedrop about once per game.

So the game has no problems there at all.

What are you talking about, you spin anything don't you .!!

1 SPE is dedicated to OS, another SPE can be used if needed but generally won't need to be (unless the game is using the eye toy or is reading HD-FMV from the blu-ray)

As for sound it is a process which is part of the game engine, same as on the 360 which uses one thread to do the sound processing.

So as Optimus pointed out the PS3 has 6 full SPE's to play with when processing in game.

SPE's are better suited to sound processing also.

4 Gigapixels / sec is 4 billion pixels per second.

1280x720 pixel resolution is equal to 921,600 pixels. Just under a million pixels.

A game running at that resolution, at 60 frames per second means the system will have to render 921,600 pixels for every frame, so...

921,600 * 60 frames per second = 5,529,600
Just over 5.5 million pixels a second for a 720p game at 60 frames per second.

That's a tiny amount given each system can push 4 billion every second. Even if the 4 billion is theoretical, the reality is still overkill.

Modern graphics call for additional pixel processing, which eats up more of that theoretical 4 billion pixels limit.

Assuming each GPU can fully tax it's memory system with pixel rendering, bandwidth writing to the frame buffer becomes the limitation.

The RSX has a max theoretical of 20.8 GB/sec writing to video memory, and an additional 32 GB/sec writing to the system's main memory.

(last time I checked it was 32 GB/sec for main RAM unless it's changed or downgraded)

Again, assuming RSX can fully tax the memory system, it can push out up to 58.2 GB/sec worth of pixel rendering to memory. (read / write)

Xenos writes to the EDRAM for frame buffer, and it is connected to it via a 32 GB/sec connection. Due to the nature of the connection this number is very close to theoretical. At first glance it seems that RSX has more bandwidth to work with, but I also see that the EDRAM logic in Xenos has 256 GB/sec bandwidth to the EDRAM memory. Again, due to the nature of the connection, that bandwidth is very close to theoretical.

So, by dividing the 256 GB/sec by the initial 32 GB/sec we see that Xenos is able to multiply it's effective bandwidth to the frame buffer by a factor of 8 when processing pixels that make use of the EDRAM functions. This includes FP10 HDR and AA and a few other tricks.

This leads to a theoretical maximum of 32 * 8 = 256 GB/sec of rendering bandwidth to the frame buffer for the bandwidth hungry tasks that the EDRAM is designed to tackle. I must say, this is an extremely effective way to brute force eliminate the performance impact of intensive effects like HDR and AA.

The main concern here is having a rendering engine that not just uses the EDRAM as a frame buffer, but takes advantage of it's massive bandwidth by putting the logic to use.

If I confused anyone as to why 32 GB/sec gets multiplied by 8, the reason is once data travels over the 32 GB/sec bus, it can be processed 8 times by the EDRAM logic to EDRAM memory at a rate of 256 GB/sec. The end result is for every 32 GB/sec you send over, 256 GB/sec gets processed. That's where the multiplication comes in.

The end result is RSX gets a theoretical limit of 52.8 GB/sec and Xenos gets 256 GB/sec.

But it is important to note that Xenos only sees that much benefit when taking full advantage of the EDRAM by putting the bulk of the graphics engine's rendering on the effects that are accelerated by the EDRAM logic.

Unfortunately for RSX, it seem that no matter how you slice it, Xenos seems to walk all over it. Not only is RSX lacking the same bandwidth saving techniques as Xenos has, but Xenos also practically eliminates the impact of heavy bandwidth effects. So not only does Xenos have a much lower bandwidth requirement, but it also has much more bandwidth to work with.

It seems clear now why Sony has been pushing the use of CELL as a means to increase the graphical output of the PS3. It seems clear that in terms of design, the RSX is not in the same class as the Xenos GPU.

This leads to a conclusion that MS made a good choice in matching the 360 Xenon CPU with the Xenos GPU. The GPU has enough power so that it not need rely on the CPU. This follows the classic design of off loading graphics work to the GPU while allowing the CPU to handle other tasks. The multi-core nature of the 360 CPU is then most benefited by it's general purpose processing.

As a result, the PS3 seems a bit of a mismatch in terms of CPU / GPU components. The system as a whole lacks general purpose processing capability while floating point processing seems redundantly placed on both the CPU and GPU. While this means that CELL has the ability to help RSX with it's strong floating point processing, RSX is still limited by rendering bandwidth. This seems to negate any real benefits of CELL for graphics, but also means general purpose processing is mostly neglected.

In addition, we also know that the 360 CPU is also capable of graphic processing. Having 3 full cores means being able to balance the needs of game, providing flexibility where needed.