Doubling the alu's sounds, fairly unlikely as they moved from 3 to 2 from Phenom to bulldozer, its possible, and its possible they are trying to introduce what would be essentially HT, seeing as the move from 3 down to 2 was said to be because Phenom rarely used the 3rd pipe making it a waste.
FPU going from 128 to 256 units is just really expected, doubling the pipelines essentially in each core(I assume it means 4 to 8 per module rather than core), is much less so but would be a huge improvement.
Still the biggest stumbling block was 4 instruction decoder shared across two cores, shared doesn't matter, a shared 8 instruction decoder would still be a massive improvement, though it seems they are going for 2x4 instruction decoder, it still theoretically doubles the number of instructions decoded per module, which will drastically increase the potential for both single and multi threaded situations.
I've said for years for AMD< Nvidia, Intel, the "first" of a new architecture almost always sucks, thats just life, we're talking about such hugely complex cores and trying to predict precisely where software will be 4-5 years in the future when you first start designing a brand new core architecture, and in particularly this architecture was to be the start of Fusion and take them well into full on APU architecture, there are massive massive massive fundamental design decisions in Bulldozer people utterly ignore. THe move towards HSA, the move towards blending gpu/cpu, and creating a software enviroment to test and optimise code in the future for that.
Steamroller always had certain things suggested for it, and it was always going to go "wider" as it went through shrinks, I said this donkeys years ago.
Intel are going less cores and wider and wider, then they will eventually move to octo cores and above, AMD went more narrower cores, and then will go wider and wider, and they'll both end up with wide/efficient/powerful architecture and 8 cores, just a different order these things happen.
As for the competition, no, competition does NOT drive prices down in CPU's or GPU's.
If Nvidia stopped making gpu's and AMD sold theres for £1000, they'd tank demand, its better to sell 30million cheaper gpu's with small profit than 2mil expensive gpu's with huge profit.
Same goes for Intel/AMD, if AMD makes the single best chip in the world, in best circumstances being VERY generous they could only forfill 20% of the worlds cpu needs yearly.... once they are out, Intel can charge whatever they want for CPU's because when 80million people need CPU's and AMD have none, you buy whatever is available. AMD making the best cpu in history would barely dent Intel sales or profits. Likewise if AMD went, it would make no difference to Intel, who currently offer worse value products in basically every segment while only offering truly massively different performance with their hexcore range..... yet they charge what they want, and they wouldn't charge anymore without AMD chips around, because they charge what the market is willing to pay.
Infact they are charging MORE than the market is willing to pay which is why their foundries have been running no where near max capacity and Intel's profits have been down. Funnily enough if Intel was making octo core chips instead of quad cores for the mainstream, they'd be bigger, and the same number of chips would take more wafers and increase capacity/reduce the waste at the foundries.
