I didn't expect it, but there's already been a lot of hype with regard to DDR3 on Intel's Bearlake chipset. ChileHardware has a nice post regarding some of the finer points regarding Super Talents announcement, as well as the details here on DailyTech.  I've had a collection of thoughts on DDR3 for some time, so I suppose it's time for a rant.

How many people got burned in the transition from DDR1 to DDR2 -- remember DDR2-400 anyone?  Even DDR2-533 provided marginal, if any gains over DDR1. 

The transitions from DDR1 to DDR2 on both platforms had some key problems.  The Intel migration to DDR2 came with architecture and platform changes that hampered its performance.  The AMD migration to DDR2 came with no performance and very little price gains.

Everybody knows it: the original Intel Prescott was a lame duck.  Mature Socket 478 setups could easily outperform Socket 775 clock-for-clock, and the Socket 775 stuff cost more.  Poor thermal solutions (at the time), a new bus that the graphics manufacturers had not fully embraced (PCIe, which also didn't enhance performance but that's another blog), and then throw in the new DDR2 memory as well ... it's a wonder the initial DDR2 Intel platforms even got off the ground with that much baggage. 

Then you had the AMD AM2 transition with mistakes that should have been learned from Intel's transition repeated.  DDR2 memory prices were considerably lower at the AM2 launch, yet there were virtually no performance gains from the AM2 platform.  However, AMD's launch was significant in the fact that it unified the industry with a memory standard -- this was the only way DDR2 prices would ever decrease.

Intel and AMD customers had very little choice at the time when their respective companies transitioned to DDR2.  This is one way to push a new format, but is it the best?

To avoid the potential hazards of a poor DDR3 launch, the industry needs to do two things, or at least one very well. The first would be to bring significant performance gains; the second would be to bring significant pricing gains.

Though benchmarks for DDR3 are still a ways out, don't expect the initial DDR3 to come close to competing with DDR2.  JEDEC is already considering expanding the DDR2 specification to DDR-1066, but the initial speed grade for DDR3 is 800 MHz.  I won't comment on where DDR3 needs to be to provide a significant advantage over DDR2, but I can tell you 800 MHz and 1066 MHz is not it.

Getting to market with low prices on any new technology is a black art.  The industry needs initial shipments to get the cost down, but no market will buy such technology without a performance advantage.  With DDR2, prices did not flatten out until after AMD and Intel had switched away from DDR1.  Historically, the only way memory manufacturers offer competitive pricing is by waiting until there is considerable demand.