engines going back to “software” renderers

Miscellaneous Forums/General Discussion/engines going back to “software” renderers

Currently CPUs are far slower than GPUs when it comes down to raw floating point calculations (which 3D rendering uses obviously) - modern video cards are pushing the "terraflop" limit (1000 GHz combined processing capability), while a current super high end overclocked supercooled CPU would be at the most 4 x 4.0 GHz (16 gigaflops).

Keep in mind that modern video cards are basically massively-multi-core CPUs with a somewhat specialized instruction set. For example my video card has about 200 cores running at a few GHz (not sure exactly the numbers). It's fully programmable - in fact some scientists and researchers are using modern video cards as miniature (cheap) supercomputers, because they're so incredibly fast at what they do.

I know that Intel is working on a new architecture which they hope will be able to compete with GPUs, but currently CPUs simply aren't nearly powerful enough to do high end 3D rendering. I'd love to think this wasn't true (since you could have so much control over rendering!), but unfortunately we're "stuck" with video cards for at least a few more years (not that that's a bad thing really - DirectX 9+ is an excellent 3D API IMO and is extremely powerful and surprisingly easy to use).