Originally Posted by nht
Except it's not faster than m.2 and SATA Express.
Maybe, it depends upon the implementation. As to SATA Express you are still implementing excess logic that impacts latency, power usage and die space. I can see the day when no SATA logic at all is included in Intels chip sets and I truly doubt we will ever see it in Apples "A" series no matter how it evolves.
The M.2 SATA Express SSDs are no slower...especially given that the Samsung XP941 seen in the 2013 MBP can be purchased as a 512GB M.2 PCIe SSD.
It isn't just the chip set. You need for the SSD to implement fast flash which few M.2 SSD do. Beyond that I'm not seeing a lot of hardware out there implementing M.2. If they are they are awfully quite about it or are selling hardware I have no interest in.
It is a shame that the MBP isn't using the M.2 standard. It would be a little bigger but probably not much.
It is a shame that Apple doesn't use some sort of standardized interface that is for sure.
There's no significant advantage for the desktop market for soldered RAM and several downsides. The stacked in-package RAM is different and separate from DRAM.
There is a huge advantage in performance with the coming new standards. Even then there are other "performance" advantages such as lower power operation. It is no surprise that cell phone processors often stack ram in the package to avoid going off chip. it isn't just about saving space.
At the high end we have this coming from Intel: http://www.extremetech.com/extreme/185007-intels-next-gen-xeon-phi-will-be-3x-faster-include-next-gen-hybrid-memory-cube-tech
. The high performance RAM stacks will be integrated right into the package. WE also have DDR4 coming which in its highest speed forms will require that the RAM be soldered in to maintain timing requirements.
It should be noted that with the coming end of easy Moore's law type improvements to CPU's, the easiest way to continue to make strides in performance is to seep up access to RAM which in some systems incurs massive delays compared to operating out of cache. Manufactures will have to work far harder on RAM interfacing to keep the performance of their machines moving forward.
Even today we are running into roadblocks where putting more cores on chip doesn't lead to the performance increases we would like to see. In many cases (certainly not al) that is directly related to problems with interfacing to RAM, the bandwidth to RAM actually.