Slashdot Mirror
DDR4 May Replace Mobile Memory For Less
Lucas123 writes "The upcoming shift from Double Data Rate 3 (DDR3) RAM to its successor, DDR4, will herald a significant boost in both memory performance and capacity for data center hardware and consumer products alike. Because of the greater density, 2X performance and lower cost, the upcoming specification and products will for the first time mean DDR may be used in mobile devices instead of LPDDR. Today, mobile devices use low-power DDR (LPDDR) memory, the current iteration of which uses 1.2v of power. While the next generation of mobile memory, LPDDR3, will further reduce that power consumption (probably by 35% to 40%), it will also likely cost 40% more than DDR4 memory."
With RAM that fast and cheap, 640kB ought to be enough for anyone!
Whoops, I mean 6.40 x 10^7 kB. THAT ought to do it.
There's a spot in User Info for World of Warcraft account names? Really?
Now we just have to wait for Intel to give a goddamn about it. Quick, somebody tell AMD to be competitive again for a few months.
I gave up talking about "need" when it comes to mobile phones long ago. It is really about "want" (for all but a very few folks who have a real need for work - most who think they "need" it for work, don't). It took me awhile to move from an old dumb phone to a smart phone. But I was finally honest with myself - and damn it I wanted one. I got one and was thrilled with all the things I can do with it. I still wholeheartedly consider smart phones a luxury - but I am glad I can afford one and finally talked myself into parting with the money and monthly payment for a data plan (I'm sort of a cheapskate). The whole family of four has them now, three of us on our second generation of them.
Go ahead and laugh. Your phone does more than I need too. But it doesn't do more than what I want.
1.2V of power??!! This is supposed to be news for nerds. Nerds should know the difference between voltage and power.
Yep, it makes you wonder why we bothered with old technology at all. Why didn't we start using today's computers fifty years ago? Think of all the time and effort it would have saved!
How can I believe you when you tell me what I don't want to hear?
In order for a spec to be useful, you need to be able to actually build the specified system. The reason they don't encompass things that they can't currently build in the specs is that they want the specs to be useful.
I'm glad they keep it reasonably simple with DDR(1), 2, 3, and now 4. I dread the arrival of RAM2015 or somesuch nonsense one day.
With Micron purchasing Elpida, Micron gonna get to make DDR4 DRAM with cell area of 4F2.
On the other hand, Samsung's DRAM is still occupying cell area of 6F2.
Muchas Gracias, Señor Edward Snowden !
After DDR5, there'll probably be DDRMAX, DDRMAX2, and DDR Extreme, if history is any indication.
If I understand correctly, LPDDR2 draws significantly less power than DDR3.
DDR4 will be competitive with LPDDR2.
But in turn, LPDDR3 will draw significantly less power than DDR4.
So manufacturers will have the choice of preserving today's mobile power levels by going with DDR4. Or they can use a more expensive LPDDR3 with lower power but, presumably, lower performance.
You think all Intel has to do is say "Hey! We'd like to support DDR4," and it just happens?
Not so much, actually. First off it has to actually, you know, be a real specification. The spec isn't final and released yet. They can't really start to use something that isn't final and subject to change.
Once it is actually out comes the harder part. They have to redesign the memory controller, which is on the chip now, to accommodate it. DDR4 isn't "DDR3 but faster," it is a different spec that works differently. Big different is no more RAM channels with multi-sticks. It is a point-to-point memory interface. So that is going to require a different setup, particularly to support large numbers of memory sticks. Also along with that the motherboards will have to be redesigned to accommodate the new RAM. Again given the point-to-point nature, the wiring would be different even if all the connectors were the same (which they aren't).
Then of course those new chips have to be fabbed, tested and made ready for sale, and those boards have to be rolled out. After all that, they still need memory. The memory manufacturers will have to retool their lines and get DDR4 chips and sticks produced in quantity to be sold.
When all that is done, then DDR4 can hit the market and go in your computer (if you purchase a new board, and processor).
So, maybe give it 6-12 months, rather than just bitching at Intel for not "giving a damn"? Just because you don't understand how something works, doesn't mean it is easy to do. Implementing a new RAM spec isn't something you just snap your fingers on, it isn't a tiny patch for software. It is a pretty major thing.
You'll probably see it in systems next year. Intel's roadmap says it will be coming to Haswell-EX server chips first, I haven't seen what AMD's plans are.
Also there's the fact that the people who post things like that are the whiny ones who had problems. I've never posted my SSD experiences before, because I'm happy, but here they are:
I have 3 256GB SSDs, one in my laptop, two in my desktop. All have worked without flaw since their purchase 11 months ago. Thus I never felt the need to go whine online about them. I've suffered no failures, no data loss. They just work.
Now, do SSDs die? Sure. So do HDDs. In terms of personal HDDs I've had about 5 fail on me over the course of my 20ish years using computers. At work, I've seen hundreds fail. Some are dead on arrival, some fail within hours of install, some fail after months or a year, some are still going strong 10+ years later.
SSDs are fine. You need to back up your data, but then that is true of anything. If you don't back up your data and have never lost anything to HDD failure that is luck, not because HDDs don't fail.
If you want an SSD the only issue should be cost. They are expensive, about $1/GB at best and as much as $3/GB for some of the really high performance/lots of write cycles stuff. HDDs are more like $0.08/GB. However if the price is acceptable, then get one. Back up the data on it to a HDD (since HDDs are cheaper, and a different technology) and you are fine. Could it die? Sure, if it does, RMA it, get a new one, and go back to what you were doing.
As others have noted, the tech wasn't there.
However, in the more abstract sense, you can only extrapolate models so far beyond the furthest point for which you have data before the models break down. But you don't know when that will happen, it depends on how good the model is and you can't know that in advance.
Specs are therefore reasonably conservative. They'll go a little bit beyond what's feasible right now, but only a little. Just enough to give wiggle-room and space for progress, but not to the point where there's a serious risk of problems developing.
Examples of reasons why they need to be careful: both electron tunneling and thermal noise will generating errors, yields at lower scales aren't always predictable, alternative techniques for performing the same function at the higher speed can be incompatible with accelerating the original technique, etc.
A spec is supposed to work for anything it is a spec for - manufacturers do revise/debug specs but they vastly prefer to release upgraded versions as compatibility issues and implementation details can all be carefully documented and properly presented.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
At least this got modded down, but it's threads like this where you discover how many ignorant 'nerds' there are on /.
Ohm's law does not only describe resistors. While the schoolboy formulation V=IR (also, admittedly, the law Ohm actually published) describes the instantaneous relationship of voltage and current through a resistor, in modern engineering and physics it is generalised in various ways. For circuit analysis, it becomes V=IZ, where Z is the complex impedance, and describes the time-varying relationship of voltage and current in resistors, capacitors, inductors and pretty much anything else you will find in a circuit. For things other than circuits, the generalisation J=E\sigma describes the relationship between electric field intensity, current density and conductivity.
While certain materials are described as 'non-Ohmic', what this really means is that \sigma is not a constant for those materials and depends on something else, usually the value of E.
Slashdot - News for Nerds, Stuff that Matters, in ISO-8859-1 Has just realised that beta makes this signature redundant