It looks like you're new here. If you want to get involved, click one of these buttons!
At CES, Micron (aka Crucial or Lexar) showed off working DDR4 memory. The advantages of DDR4 over DDR3 are pretty straightforward: higher clock speeds at lower voltages. The latter means lower power consumption, and the difference between 1.2 V and 1.5 V at stock voltages should mean 36% less power consumption for DDR4 if all else were equal (which it isn't). DDR4 will start at 2133 MHz (real clock speed of 1066 MHz, then doubled because memory sellers like to claim higher numbers) and clock speeds will go up from there.
DDR4 has been greatly delayed, however. For example, AMD's Vishera chips that are out now (FX-*300) were supposed to be paired with DDR4 memory. Or rather, it was some other chip that was supposed to use Piledriver cores paired with DDR4 memory and Vishera wasn't supposed to exist, but DDR4 wasn't anywhere near ready, so AMD canceled the planned new platform and just fixed up the cores while fitting an old platform.
It's also interesting that Micron is saying that DDR4 will first be used in servers. Power consumption is a huge deal in data centers, and that's the area most willing to pay a premium price for reduced power consumption. But that leads me to wonder which chips will be the ones that use DDR4. Most Intel and AMD server chips are also used in desktops and/or laptops. But those chips usually launch for desktops and laptops first at the same time as servers if not earlier, as servers have higher reliability demands that require a longer validation process. If you make a chip with a DDR4 memory controller, then you can't just plug DDR3 into that and expect it to work.
The notable exception to x86 server chips also appearing in desktops and laptops is Intel's super high end whatever-EX platforms. But I don't think that DDR4 makes sense there, as those are for servers that need a huge amount of performance in a single server with coherent memory and so forth, and that burns a lot of power.
This is my speculation, but I think the most logical candidate is microservers, with many low-power cores. That's the product targeted at places that are very sensitive to power consumption. Intel's Avoton Atom-based servers would make sense. So might the promised Jaguar-based Opteron chips that AMD is promising. I don't think that the ARM v8 server chips will be ready by the end of the year, but Calxeda, Marvell, and so forth will probably have ARM v7-based servers soon. (Same ISA as Cortex A7 and A15.)
DDR4 for desktops and laptops will get there when it gets there, and in many computers, the extra bandwidth doesn't really matter. Consumers may wish to shy away from the higher initial prices, too. But eventually, DDR4 will be a big deal in APUs, as if you're trying to feed integrated graphics, you need all of the memory bandwidth that you can get. Starting at 2133 MHz and going up from there means you'll have a lot more bandwidth than DDR3 offers, and there should be much higher binned modules available early on.
Micron says that the initial DDR4 modules will be on a 32 nm process node. That surely means they'll have a large die size and be expensive, though they'll do a die shrink as soon as they can. My guess is that Micron wanted a mature process node so that they weren't having to fight with process node problems when trying to get the hang of DDR4 production down.