Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Behold the incredible, shrinking XPoint performance!
In mid 2015, Intel and Micron announced that they were working on a revolutionary storage technology that they called 3D XPoint. Intel has since called their version Optane, while Micron has called theirs QuantX.
http://www.anandtech.com/show/9470/intel-and-micron-announce-3d-xpoint-nonvolatile-memory-technology-1000x-higher-performance-endurance-than-nand
Look at the claims there: 1000x faster than NAND, 1000x the endurance of NAND, 10x denser than DRAM. Impressive, no?
Sometime after that, reality got in contact with Intel and forced them to scale back their claims, and repeatedly. The first Optane products for consumers are supposedly going to be some stupid cache that won't change much other than the price tag, at least assuming it's used in an otherwise sensible rig that has an SSD, which it probably won't be. Now there is a claimed leak of specifications for the first enterprise product:
http://www.tomshardware.com/news/optane-3d-xpoint-intel-p4800x-cold-stream,33624.html
Sequential read and write performance are comparable to enterprise NAND-based SSDs. So is random read performance. Random write performance is substantially, and it can supposedly do that at lower queue depths, too. But we've now shrunk from 1000x faster than NAND to something less than 10x in random write heavy workloads, and closer to 1x in most other workloads.
The latency advantage has mostly vanished, too, going from tens of nanoseconds all the way to ten microseconds. That's better than NAND, but it's not even 10x, let alone 1000x.
Endurance is up, too, but we're looking at maybe 200x as compared to cheap consumer SSDs, or 20x as compared to enterprise NAND-based SSDs. That's considerable, but it's a whole lot less than 1000x.
And you know what we're not talking about? The price tag. It will probably be cheaper on a per-GB basis than DRAM, but the assumption all along has been that it would be a lot more expensive than NAND. And now we're looking at a lot more expensive than NAND with few meaningful benefits over NAND.
It's quite possible that it will still be a nifty product for certain niche, enterprise uses. But there will surely be a whole lot fewer such use cases where it makes sense than if the specs were closer to the initial hype.
http://www.anandtech.com/show/9470/intel-and-micron-announce-3d-xpoint-nonvolatile-memory-technology-1000x-higher-performance-endurance-than-nand
Look at the claims there: 1000x faster than NAND, 1000x the endurance of NAND, 10x denser than DRAM. Impressive, no?
Sometime after that, reality got in contact with Intel and forced them to scale back their claims, and repeatedly. The first Optane products for consumers are supposedly going to be some stupid cache that won't change much other than the price tag, at least assuming it's used in an otherwise sensible rig that has an SSD, which it probably won't be. Now there is a claimed leak of specifications for the first enterprise product:
http://www.tomshardware.com/news/optane-3d-xpoint-intel-p4800x-cold-stream,33624.html
Sequential read and write performance are comparable to enterprise NAND-based SSDs. So is random read performance. Random write performance is substantially, and it can supposedly do that at lower queue depths, too. But we've now shrunk from 1000x faster than NAND to something less than 10x in random write heavy workloads, and closer to 1x in most other workloads.
The latency advantage has mostly vanished, too, going from tens of nanoseconds all the way to ten microseconds. That's better than NAND, but it's not even 10x, let alone 1000x.
Endurance is up, too, but we're looking at maybe 200x as compared to cheap consumer SSDs, or 20x as compared to enterprise NAND-based SSDs. That's considerable, but it's a whole lot less than 1000x.
And you know what we're not talking about? The price tag. It will probably be cheaper on a per-GB basis than DRAM, but the assumption all along has been that it would be a lot more expensive than NAND. And now we're looking at a lot more expensive than NAND with few meaningful benefits over NAND.
It's quite possible that it will still be a nifty product for certain niche, enterprise uses. But there will surely be a whole lot fewer such use cases where it makes sense than if the specs were closer to the initial hype.
Comments
The big advantage on 3d NAND is that its gotten to very acceptable $/GB levels. Just look at how much 16 GB of RAM costs and its $/GB compared to NAND to get idea of how much Intels offering of 16/32 GB cache will cost (it wil be cheaper than DRAM but its Intel, and idiots will buy it anyway)
It's not looking so good though.
Recalling that story that was linked here a few days ago (where Intel is moving away from Desktop PCs and more towards the data center), they had a chart that showed Intels' own revenue estimates going forward, and Optane growth played a very prominante place in that. Intel is betting big on Optane growth
Now, maybe they can take the Microsoft approach - where you can release an otherwise mediocre product, and continue to support, refine, iterate, and promote the product until it becomes the dominant force. Intel hasn't been very successful with that tactic though, if their mobile strategy is anything to go by.