AMD FX "Vishera" Processor Pricing Revealed

[inf64]

At XS AMD section,a poster already gave us Vray numbers in real world scene rendering. To save you the time , new FX is ~10% faster than the old at the same clock. And it can clock a bit higher on air/water. For a same node and what was supposed to be a "quick fix of Bulldozer", improvement seems pretty good.

edit: x264 numbers. ~10% faster than old FX at same clock.
3dstudio scene rendering time of FX8350@ 4.5Ghz vs FX8120 @ 4.5Ghz and 2600K @ 4.5Ghz. FX8350 is 7% faster than 2600K and 15% faster than the old FX, all tested on 4.5Ghz fixed clock.

 

[kelco]

So what you're sayin is they rushed bulldozer?

 

[inf64]

Are you referring to me? No ,why would I say that?

 

[Dresdenboy]

So what you're sayin is they rushed bulldozer?

Bulldozer was late (original plan: 2009 in 45 nm). Different reasons likely held them back (pick your favourites): power consumption, die size, design maturity, budget, personnel, complexity etc.

So far it looks like a staged approach (2011: BD, 2012: PD, 2013: SR, 2014: EC):
- have a list of features to be added to a µArch
- add a subset of that per stage as resources permit

Since there is a 4+ years cadence of µArch development, it would be difficult to prove that BD results led to changes in the succeeding architectures. If some improvement appears in later designs, it is either caused by a delay or by design.

IIRC, Bulldozer's chief designer Chuck Moore (R.I.P.) once said, that he would only allow microarchitectural features to get into the design if they a) increase performance more than power consumption and b) are not overly complex (as this could need too much time for design, development, test&validation).