blazorthon
Glorious
[citation][nom]moricon[/nom]Global Foundries WAS AMD !!! http://en.wikipedia.org/wiki/GlobalFoundriesNow AMD will have a little more flexibility in deciding who gets tender of Manufacture on their design. Of course AMD have to pay to get out of exclusivity contract, GF was essentially SOLD in that state to private investment Consortium with Abu Dhabi's Advanced Technology Investment Co! being the largest.They would have wanted AMD tied to using them for a period of time, or what would be the point in them investing!This still leaves AMD short on their Design front though, and that is where AMD has real issues in the CPU world.. Maybe the CPU design guys should spend more time with the GPU guys to see how to execute correctly![/citation]
To be fair, the GPU guys in AMD seem to have failed to understand the basics of performance scaling in video cards and their GPUs. IE 7870 is almost identical to 7950 in performance because AMD gave it a 1GHz clock rate instead of an 850MHz or 900MHz or giving it more shaders.
1280 shaders @1GHz is right behind 1792 shaders @ 800MHz because clock rates scale more linearly than having more shaders. Then there is the 7850 that isn't even stable, and GCN Radeons don't have Crossfire support yet, there's the inconsistencies in performance (sometimes the 7870 performs about as well as a GTX 580, some times it's beaten by the 560 TI and the Radeon 6950). We also have 7900 and 7800 having the same number of the more or less same ROPs as 6900. I can understand 7800 doing this, but 7900 should either have improved ROPs or more of them.
A lot of Bulldozer's problems aren't only the people, but how it was designed. For example, some performance critical parts of a CPU are hand optimized. Bulldozer is completely computer designed instead, the lead engineer at the time said that this made it 20% larger, use 20% more power and be about 20% slower. He also said that he'd tested computer designs before and that this is a relatively constant problem in all purely computer generated designs. If this alone was fixed, well that would push FX above Phenom II and probably into the realm of Nehalem, making it at least better than it is now and better than it's predecessor. Then we have AMD needing to fix it's low performance cache (latencies are huge on all AMD cache, but FX is again, worse than Phenom II in this regard too).
After that, well then it's it's design that is the problem. Some things were simply skimped on that shouldn't have been. However, if AMD gets Piledriver at least up to Nehalem in performance per core, then those 8 core chips should beat even the quad core i7s in highly threaded work, so AMD could finally have a performance win if they price them like an i7 or cheaper. AMD would need to make some serious improvements to catch up to SB or IB in IPC, but I think just passing by Nehalem should be enough of a performance increase for me to consider buying FX for highly threaded work. Think about it, just fixing the auto designed parts means it could be between Nehalem and Phenom II in performance per core and power usage, a fair improvement.
To be fair, the GPU guys in AMD seem to have failed to understand the basics of performance scaling in video cards and their GPUs. IE 7870 is almost identical to 7950 in performance because AMD gave it a 1GHz clock rate instead of an 850MHz or 900MHz or giving it more shaders.
1280 shaders @1GHz is right behind 1792 shaders @ 800MHz because clock rates scale more linearly than having more shaders. Then there is the 7850 that isn't even stable, and GCN Radeons don't have Crossfire support yet, there's the inconsistencies in performance (sometimes the 7870 performs about as well as a GTX 580, some times it's beaten by the 560 TI and the Radeon 6950). We also have 7900 and 7800 having the same number of the more or less same ROPs as 6900. I can understand 7800 doing this, but 7900 should either have improved ROPs or more of them.
A lot of Bulldozer's problems aren't only the people, but how it was designed. For example, some performance critical parts of a CPU are hand optimized. Bulldozer is completely computer designed instead, the lead engineer at the time said that this made it 20% larger, use 20% more power and be about 20% slower. He also said that he'd tested computer designs before and that this is a relatively constant problem in all purely computer generated designs. If this alone was fixed, well that would push FX above Phenom II and probably into the realm of Nehalem, making it at least better than it is now and better than it's predecessor. Then we have AMD needing to fix it's low performance cache (latencies are huge on all AMD cache, but FX is again, worse than Phenom II in this regard too).
After that, well then it's it's design that is the problem. Some things were simply skimped on that shouldn't have been. However, if AMD gets Piledriver at least up to Nehalem in performance per core, then those 8 core chips should beat even the quad core i7s in highly threaded work, so AMD could finally have a performance win if they price them like an i7 or cheaper. AMD would need to make some serious improvements to catch up to SB or IB in IPC, but I think just passing by Nehalem should be enough of a performance increase for me to consider buying FX for highly threaded work. Think about it, just fixing the auto designed parts means it could be between Nehalem and Phenom II in performance per core and power usage, a fair improvement.