Is Broadwell toast?

Computer chip advances have long been driven by Moore's Law.  Periodic die shrinks meant more transistors available in the same space, and that enables performance to go ever upward.  While architectural advances have played an important role, too, die shrinks that enabled new architectures to have more transistors available were indispensable in enabling improved architectures.  As recently as 2011, with the launch of Intel's Sandy Bridge architecture that handily bested all previous architectures and AMD's Bulldozer on the horizon, it seemed that such improvements could continue for years to come.

But then a funny thing happened.  Ivy Bridge was a die shrink of Sandy Bridge from 32 nm to 22 nm.  As one would hope, it was faster at stock speeds while using less power.  But it couldn't overclock as far as Sandy, so if you overclocked both, it was essentially a tie.  Haswell launched earlier this month, and while it wasn't another die shrink, it did repeat Ivy Bridge's path:  a little faster at stock speeds, but can't overclock as far, so if you overclock them all, Haswell wasn't really any faster than Sandy or Ivy Bridges.

On the AMD side, things were even worse.  AMD hadn't caught up to 2009's Lynnfield before Intel launched Sandy Bridge, and Bulldozer didn't do much to help, as it wasn't really any faster than AMD's old Phenom II chips.  In spite of being a full node die shrink from 45 nm to 32 nm, Bulldozer wasn't any more power efficient than Thuban.  While Bulldozer cores were rather broken and Piledriver cores largely fixed them, performance still trailed far behind Intel.  AMD has Steamroller cores coming on a 28 nm process node later this year, and those should be markedly better than anything AMD has on the market today, but they probably still won't catch Intel's Sandy Bridge.

But never fear:  Intel has Broadwell on a 14 nm process node coming next year.  Then came rumors that Broadwell wouldn't release for desktops at all.  Or perhaps that it would release, but not with a socketed version.  Or maybe there would be a socketed version, but only one, and mostly it would be BGA versions for low power systems.  None of that would make sense if Broadwell was going to be faster than Haswell.  To be fair, higher desktop performance isn't the only goal; Haswell could still be a nifty chip for laptops because it brings power consumption way down.

Then came recent leaks:

http://www.dailytech.com/Report+Intel+Delays+14+nm+Broadwell+Schedules+Haswell+Refresh+for+2014/article31770.htm

That's Intel's desktop roadmap through 2014.  Notably missing is Broadwell.  Notably present is Atom, Intel's very low end chip that has never had a socketed version, so this isn't just a case of only showing the high end chips.  But while 32 nm Valley View appears, it appears in 2014.  The 14 nm Airmont chip that was promised for 2014 is completely missing.  Also notably present is a Haswell refresh, which wouldn't happen if a superior Broadwell chip were set to appear mere months later.

So what happened to Broadwell?  Is it canceled entirely?  Delayed until 2015?  Is it really going to be laptop-only, with even the semi-defective chips that require too much voltage not diverted off into desktops or other form factors that are more tolerant of higher power consumption?

And who would have guessed that four years after the launch of Sandy Bridge, it would still be competitive with the latest and greatest high end chips?