I'm fairly new to chip architecture as well but I think what your saying sounds right. Although the PS3 has partitioned RAM (2x256) opposed to the 360's shared RAM (512) I believe the RSX can use both sets of RAM so it's difficult to know what the RAM restrictions of each GPU really are with these 2 designs. Also OS RAM usage needs to be taken into account.

I get the feeling that PS3 exclusive developers can optimise a game for the RSX as they know how many pixel/vertex shaders they have available and can adjust the engine accordingly. Of course they can also use the Cell processor to offload as well. However, for multi-plat developers this is a problem and the unified shader architecture of the Xenos is much easier to program for and far more efficient.

The reason why the unified shaders are a good approach is that most games have a different workload of pixel/vertex shaders than the GPU designer imagined which results in one acting as a bottleneck and lowering real-world performance of the game on the system. If you have a unified shader architecture your system has much higher performance in either extreme, but also obtains a much more natural balance with fewer performance bottlenecks in most situations.

Marking this thread BRB

The Xenos is based on ATI's R600 architecture, which is substantially newer than Nvidia's G70. Nvidia's next-gen chip, the G80, was the one which competed with R600 on the desktop and that has unified shaders too.

Incidentally the Wii graphics chip is a heavily optimised R300 which is years old.

So, yes, the Xenos is more powerful and flexible. A potential "PS4" would use a graphics chip with unified shaders, since both ATI and Nvidia use them and they are the only choices at the momemt. Unless we get Intel's Larrabee.

How can it be a R300 when its features aren't similar at all to the R300? I can't remember the details, but my vague memory tells me they haven't got much in common.

@Thread:

Lets see:

The Xenos has better shaders and shader architecture. It can run more complicated shaders more efficiently, it has a unified shader architecture, it can run 3 different threads at once IIRC, it can write directly to the L2 cache of the Xenon CPU, it has 2* the MSAA sampling rate of the RSX, high memory bandwidth, doesn't pay the memory cost of MSAA writebacks to main memory, massive fillrate, it can perform some GPGPU tasks for the CPU as well, (Saw a mention somewhere in a MS doc), has a built in tessellator (Not really used unfortunately as its kinda limited). Thats about all I can think of, theres probably quite a few more.

Xenos also has daughter die, which is used to crunch some pixel shader stuff. Thats where the 'free' AA comes. :P

http://en.wikipedia.org/wiki/Xbox_360_hardware

^ Hehe, I forgot about those...

Btw I saw in one of the performance docs that the GPU was actually CPU limited. It won't be until can squeeze a bit more power out of the CPU as the programming matures that we will see the limits of the GPU.

Just as the Xenos is an early R600 prototype, the Gamecube's "Flipper" GPU was designed by ArtX, which ATI bought and used their IP to make the R300. So it's not an R300 but it's the closest in architecture to it.

I see where you're coming from. Btw the Xenos and RV770 Gpus were designed by the same team, R600 was done by their other team. Vaguely, something in my head says that the ArtX team was the basis of the Xenos/RV770 team as well, but im only 30% sure on this.

For the first part: Link?

Well, ArtX got bought out quite a while ago. I don't think you can point to one part of AMD and say: "that's the ArtX team". All of their GPUs since R300 have been heavily based on ArtX technology.

Unrelated; how long to you think people will refer to ATI as ATI? Now that it's AMD?