Originally Posted by brucethemoose
The proprietary DIMM drives will probably be faster than the PCIe ones though. And I don't see Intel making the XPoint DIMMS compatible with AMD CPUs anytime soon... Althought that will almost certainly be an enterprise only thing anyway.
Why would they not? They aren't competing against AMD in the SSD market, after all. And it isn't just Intel but also Micron who is selling these thanks to their partnership. It's sort of like Apple using Samsung components in iPhones. Sure, Samsung might have produced it, but it's not like Apple has fabs to make their own RAM or a department working on their own SSDs. Companies in this industry can't just be watered down to a name and a logo since all of them are making so many different things.
Originally Posted by GANDALFtheGREY
Still really interested to see the DIMM implementation of these chips tho and what can be done with this type of multi purpose memory.
That concept already exists, sort of
. The goal is to both minimize latency and make it extremely predictable rather than maximize bandwidth. But ULLtraDIMMs or however it's capitalized aren't allowed to be sold in the US anymore thanks to a lawsuit
. It isn't a "patent troll" type lawsuit either and seems like a pretty valid excuse to go to court. Unfortunately it also impacts SanDisk (and indirectly their server-making customers such as HP and IBM, and business which buy those
servers...) and the progress of some nifty tech.
Presumably 3D Xpoint will incorporate the same ideas and be better overall. NAND flash and NOR flash (NOR flash being much slower and often less durable, but accessible at the bit level) are very old, introduced during the '80s, and have some major flaws inherent to them such as erasing and durability. The way modern flash drives are accessed also adds a ton of latency, going from the CPU to the southbridge to the SATA controller to the SSD controller and back again. SATA and even PCIe based flash is incredibly high latency, taking thousands of clock cycles to access, compared to system memory at hundreds, last-level cache (LLC) at dozens, and L1 cache at <5. (I think, I can't remember the exact values and might be off by a couple zeroes.)