Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
3D XPoint
http://newsroom.intel.com/docs/DOC-6713
Does NVMe suddenly make more sense?
This obviously isn't shipping, and there's no indication of when it will actually show up in a product (let alone what the price would be) - but it could seriously shake up storage if even half of what they print here is true.
Comments
I don't know, SSDs are still (much more than) 4x as much as HDD for a given capacity, and they are going pretty well.
And when SSDs first came into the consumer realm, they were significantly more expensive than they are now, and they still caught on and are well on their way to being mainstream (there are some who would debate that I suppose, but I think most wouldn't).
SSDs had large and obvious advantages over hard drives as soon as Intel demonstrated that you needed good random write performance.
Will XPoint have large and obvious advantages over SSDs for consumer use? I'm skeptical of that. For SSDs, if this SSD is twice as fast as that one, it doesn't matter for consumer use, as they're both plenty fast enough.
I certainly believe that software of interest to consumers could be written to take advantage of XPoint and perform better with XPoint than with SSDs. For enterprise use, in some cases, you'll do exactly that, as you can assume that you have whatever hardware you have, or require a customer to buy expensive hardware in order to use your expensive software. But people aren't going to write such consumer software to take advantage of XPoint if there aren't a ton of XPoint devices already in consumer use.
Well of course they're going to patent it. It would be irresponsible not to. You patent anything and everything first and then work out the licensing agreements later.
It's also important that, as this was developed by IMFT and not Intel, both Intel and Micron have access to it.
SSDs had large and obvious advantages over hard drives as soon as Intel demonstrated that you needed good random write performance.
Will XPoint have large and obvious advantages over SSDs for consumer use? I'm skeptical of that. For SSDs, if this SSD is twice as fast as that one, it doesn't matter for consumer use, as they're both plenty fast enough.
I certainly believe that software of interest to consumers could be written to take advantage of XPoint and perform better with XPoint than with SSDs. For enterprise use, in some cases, you'll do exactly that, as you can assume that you have whatever hardware you have, or require a customer to buy expensive hardware in order to use your expensive software. But people aren't going to write such consumer software to take advantage of XPoint if there aren't a ton of XPoint devices already in consumer use.
Presumably (because we don't have shipping products yet)...
It's not twice as fast as existing SSDs.
It's several orders of magnitude faster.
That's the kind of change that disrupts tech.
It doesn't have the speed or endurance to be used as DRAM. It's probably too expensive to be used for mass storage. It could be nifty for virtual memory; 8 GB of DDR4 plus 32 GB of XPoint might plausibly work as well as 32 GB of DDR4 while being cheaper. But there aren't many consumer applications that need a ton of system memory.
I could certainly see software that knows you have a bunch of XPoint using it to great effect for things that SSDs don't work for. But I'm very skeptical that there's a lot of such software just floating around out there now in broad consumer use already--and I'm also skeptical that anyone will write such software for consumer use that relies on hardware that virtually no one has. When SSDs arrived, there was already a ton of software in wide use that benefited from the speed of an SSD over a hard drive.
As I said above, that's not a problem for some enterprise uses. In some cases, you choose the hardware and write the software together, so you can write software to take advantage of whatever hardware you decide to get. That avoids the "chicken and egg" problem that is inevitable for consumer use.
Your making an assumption on the price and basing it's unsuitability toward consumer everything on that.
That ~may~ be true. It also may not be true.
I'll agree with that. I'm very skeptical that initial products will be priced competitively with consumer NAND SSDs, but wouldn't be terribly surprised if that changes inside of five years. Even enterprise NAND SSDs aren't priced competitively with consumer SSDs.
The marketing hype says 1000x as fast as NAND. It's unlikely that that means bandwidth; if it did, that would wipe out not just DDR3/DDR4 SDRAM, but also GDDR5 and HBM. If it means 1000x less latency, then that puts the latency on the order of 100 ns. That's far too slow for your main system memory; it would be like buying DDR3 or DDR4 with latency specs in the low hundreds.
For endurance, it also says 1000x the endurance of NAND. That would mean you can write over it millions of times before wearing it out. With DRAM, you could burn through that many write cycles in mere seconds if you try to. Even if your writes are spread out evenly across the entire memory space, some GPUs could do 10 million writes over their entire memory inside of a week. Some GPUs probably do do tens of millions of writes across most of their memory space in a typical year.
XPoint might well be used to great effect as an intermediate level of cache between DRAM and SSDs. But in no way does it mean you can get away without having some DRAM for system memory.