Will 512-bit bus + GDDR5 memory have a place in 2009 ?

[Mintmaster]

What if they had debuted with a 55nm 128 shader chip with G200 architectural improvements, a 384 bit GDDR3 bus and 2 GHz shader clock when the G200 was supposed to have hit the market?

Could they fit a 384-bit bus on there if it was < 300 mm2? Do GT200 improvements help their perf/mm2?

I don't think this would be more than 15% faster than G92b, and it would cost a lot more (50% more RAM and larger die size). G92 already was top dog two quarters ago, so I don't see this chip really helping them much.

Moreover, I think RV770 would still come out on top at $299.

 

[MfA]

Could they fit a 384-bit bus on there if it was < 300 mm2?

I think so, yes.

Do GT200 improvements help their perf/mm2?

It would certainly help their FLOPs.

Moreover, I think RV770 would still come out on top at $299.

They would have had been reaming high margins for 6 months before the RV770 ever hit, they could have priced it higher than the G92b and still sell them (they could have hit significantly higher clocks than the G92, simply because they could have designed it for significantly higher clocks). They would have been ready to move to a bigger monolithic at the same time they are doing it now for the G200b.

They wasted a huge amount of time on the G200 ... and there is only so much work they can do in advance. Anything would have been better than sitting on a core for half a year, hitting 55nm at the same time as ATI with a "simple" core would have been better.

 

[CarstenS]

Do GT200 improvements help their perf/mm2?

Not necessarily wrt to peak rates (i.e. marketing), but IMO quite a bit in terms of achieved utilization (i.e. usefullness).