More PlayStation 3 Grid Computing Details

gwernol writes: "Redherring has a good article on Sony's forthcoming PlayStation 3: not too many technical details but good background to the Xbox/PlayStation wars. Sony are touting the use of massively parallel 'cell computing' to get a 1,000 times performance increase over the PS2. This plan, also known as grid computing is also discussed here."

Comments

Sounds a bit "iffy", sir

[Space cowboy]

Hmm. That's very odd.

The PS2's geometry engine is a 4x4 arrangement (16 pixel pipelines in total), so the fastest possible render is irrespective of 4x4 or 0x0. Given that a 4x4 triangle at least possesses triangular nature, I'm surprised that they would go for 0x0.

The PS2 also doesn't use co-ordinate space of 0,0 to be anything special - the hardware has automatic scaling from an abstract 4096x4096 space into whatever resolution you happen to be working in. Typically (at 640x480), the co-ordinate (0,0) is at (1728,1808). Why then does it matter where you render the triangle ?

Whether you draw the buffer to the screen or not is also not relevant - you can't "draw" to local memory - it's drawn to RAM with a 2048-bit bus on the same chip as the video processor. There's no reason why displaying the screen would slow that down, so why open yourself to criticism if that were the case. Odd behaviour to say the least...

In short, I think you're wrong.

There are detailed figures for different types of draw operation in the GS users manual. The 75 million/sec is for no-texture, flat-shade, no-anti-alias. It drops down to 16 million/sec for textured, gourard shaded, fogged, anti-aliased triangles.

Simon