Linked by Thom Holwerda on Mon 8th Jan 2007 20:54 UTC, submitted by anonymous
Thread beginning with comment 200025
To view parent comment, click here.
To read all comments associated with this story, please click here.
To view parent comment, click here.
To read all comments associated with this story, please click here.





Member since:
2005-07-06
This isn't even close to true. Cell is far from being ideal for rasterizing 3D graphics. Take something very simple like trilinear filtering. For every pixel you write out, you need to load 8 texels and blend them together. The SPE can do one memory load and one arithmetic operation per clock cycle. You're talking dozens of clock cycles just to get a single texel value, not to mention the final blending or multi-texturing or bump-mapping, or any of those things. Meanwhile, GPUs have multi-ported texture caches that can do the memory reads in just a couple of clock cycles, and lots of dedicated interpolation hardware in the texture units that can allow the chip to provide a texel to the pipeline every clock cycle.
There is a reason GPUs are advertised as performing trillions of operations per second. They have a ton of fixed-function hardware that is necessary for the general-purpose units (the shaders) to do their work efficiently.