ACube Systems

[Hammer]

As for gpu's, we know the GC had done 15 million polygons/second in real time

M9 can do 40 to 45 million triangles (polygons)/second and provides hardware-accelerated support for DirectX 8.1 class shader model.

M9 has support for QXGA resolutions (2048x1536) (i.e. beyond 1080p).

[Hammer]

As for the performance of the ATI mobility gpu, mobile chips always lagged behind in laptops

So? Given similar GPU generation they don't lag to far behind in performance i.e. they can still post unplayable fast frame rates scores in game benchmarking.

is there a real game where this was achieved or is it a test like MS's claim that the original XBOX could do 130 million but in-game was only 18 million at best.

NV2A based architecture was use in higher performing Geforce 4 TI.  

The GPU has it's own bank of 16MB of memory and a 3MB internal texture cache and direct access to the disc drive, so it could be streaming audio and graphics data from disc INDEPENDENTLY of the cpu.

Recall the purpose of "AGP Fast writes" and DMA protocol.

Intel P6 FSB 64bit FSB in XBOX1 doesn't consume nForce’s 128bit wide bandwidth.

The GC's cpu had more cache (and programmable at that) than either the Xbox

Recall NVIDIA's DASP function in nForce chipsets
http://en.wikipedia.org/wiki/NForce

DASP amounts to 3rd level cache functions.