Soleron said: @Squillam |
Its not AMD who are below fab capacity its Global Foundries, just a quick correction. However the concept that they have fab space to give away may very well be coming to an abrupt end once they bring their graphics on board in both Fusion which will be a larger higher margin core with lower yields and their standalone GPUs as well as the global partners who want to use their high quality fabrication processes as TSMC has dropped the ball as of late.
The performance requirements of all console -> PC ports have been increasing with time as developers make better utilization of the Xbox 360 CPU. People used to say that you needed a single core to match it, then it was dual core and now a lot of games are starting to recomend triple/quad cores for games which were ports from the console. The ports aren't getting worse, the demands from the games are increasing. The Xbox 360 CPU has been specialised somewhat for games and out of order execution is of less benefit when code is directly compiled for a CPU itself.
IBM themselves don't have a huge desktop market to fall back on so they take royalties for development where they can. This is why they are more ameniable to console manufacturers because they need them from a software development and research and development standpoint. X86 doesn't need any more developers than it currently has. They also don't need to compete with themselves by giving away cheap mini computer components to compete with their low end desktop lines.
Microsoft on AMD extensions yep I see where you're coming from
Please don't write off Fermi before its come out. Its a pretty big leap to suggest it would only perform ~20% faster than Cypress when it hasn't been released yet. Theoretical FP performance is just that, theoretical. Theoretically a non unified PS3 GPU has more flops than a unified Xbox 360 GPU but in reality the latter outperforms the former. Whilst Cypress is a great GPU its an architecture which has been scaled from 320 stream processors to 1600 stream processors, from 600M transistors to 2,000M. So if the Fermi architecture is a good one it should be able to easily beat Cypress on effective flops and it definately has an advantage in flops vs bandwidth due to a wider bus.
Tease.