Oh I understand that this doesn't include AVX2 or Tsx . But thats a wait and see , Sure it helps make haswell future proof . I not going to go the way of others point at future apps that will use next generation apps and recompiles. Of course I would like to see intel 8 real cores against the AMD processor with 8 cores . That is no longer how we measure performance . We now measure performance based on price. Because No one can come close to 8 real intel cores. It was so easy in 2006 . I new this game was going to played this way . As I stated long befor it happened in real life . Intel played its core hand vary well . That doesn't change the fact I disappointed in only 10% gain in that benchmark . As I was expecting HT performance improving . With this benchmark I don't know where the improvement is as it uses HT.
This should be somewhere else but how much of Intel's success is their own competence, and how much is AMD's lack thereof?
When I do a SWOT analysis, I see Fab leadership, brand, marketpower, processor performance leadership, profitability, reliability and ecosystem/chipset. Weaknesses I see high Fixed Costs, waste, excessive segmentation and inferior GPU technology.
For AMD, their strength is GPU technology, their weakness is everything else but particularly, their weak financial disposition, their generations behind fab technology, the WSA burden, and their me too strategy.
Now, what happens if somehow AMD were to get rid of the WSA, strengthen their balance sheet, outsource to TSMC (which while still behind Intel, is much lower cost), stop segmenting and change their product strategy from me too to cross subsidization based on their strengths?* They'd be a formidable competitor, even with inferior processor technology.
*to break-even, AMD only needs to make ~$20 profit per CPU, regardless of price or socket. So if for example, their servers shared the same chipset and socket as desktop, they could probably sell a Server CPU for $50 and still make money - its the Southwest Airlines vs the Big Airlines model, point to point/flat pricing vs hub and spoke/revenue management.
** they could push the model further and price everything based on customers buying a CPU/GPU bundle (or APU) in which case the needed profit is more like $35 per bundle and with 1 segmentation between APU and CPU/GPU (ex. $20 profit x 10MM APUs + $60 profit x 5MM CPU/GPU bundle) which would make the CPU very inexpensive if a customer bought a 7870 or higher end card - possibly free if the video card and motherboard were combined into 1 circuit board. Which by the way, could cause a mini-boom in the x86/PC industry - big improvements in price/performance usually does that.
We know this will never happen - this is AMD after all - but what if it did?
Don't be so confident - only the paranoid survive.
Facebook
Twitter