Intel Promises 'Optane' SSDs Based On Technology Faster Than Flash In 2016 80
holy_calamity writes: Intel today announced that it will introduce SSDs based on a new non-volatile memory that is significantly faster than flash in 2016. A prototype was shown operating at around seven times as fast as a high-end SSD available today. Intel's new 3D Xpoint memory technology was developed in collaboration with Micron and is said to be capable of operating as much as 1000 times faster than flash. Scant details have been released, but the technology has similarities with the RRAM and memristor technologies being persued by other companies.
But, but... (Score:4, Funny)
He's the fastest man alive! Ain't nobody faster than the Flash! [wikipedia.org]
They lie...
Re:But, but... (Score:4, Funny)
Price point? (Score:3)
Re: (Score:3)
Re: (Score:2)
That pricing curve does not remain consistent. Sometimes in consolidating markets it is much better to exploit a technological lead by coming it at a low price in order to cripple your competitors. So push SSD drives to completely wipe out H/Disk drives and hurt the manufacturers bound to H/Disk drives.
Re:Price point? (Score:5, Interesting)
I read the TFA for a change and it said that the drives would be available in lightweight laptops. There was a Slashdot article about this technology when it was announced and there was speculation about whether it was a cache technology or a direct storage medium.
TFA said it would be for enterprise storage *and* laptops, so its likely means it's a "drive" and that it will likely be more or less affordable or it wouldn't go into laptops. The only question is how fast, TFA said the demo was only 7x current flash but maybe faster at introduction in 2016.
I think the original story said it was far more durable than flash now, and if TFA article is to be believed about use in enterprise storage it could really shake things up. Vendors now make a big deal out of fancy tiering schemes, charging two arms and a leg for a few SSDs and fancy software to keep quiet data on their only slightly cheaper nearline disks. What's the point if you can do 100% of this with drives faster than flash but cheap enough to go into laptops?
Re: (Score:1)
I personally pretend a technology doesn't exist until I see it widely available in the channel with the various price points. Since the tech is going to go to DIMMs as well, I suspect the price will be high and the main usage at first in caches and buffers (burst buffers in HPC, for example).
Re: (Score:2)
I personally pretend a technology doesn't exist until I see it widely available in the channel with the various price points.
remind me not to hire you for budget forecasting
Re: (Score:2)
Haven't we already got to the point where 'fast' isn't really a problem any more?
Shouldn't we be looking more at things like 'reliable' and 'price'?
Re: (Score:2)
For consumers, price, speed and reliability are pretty much solved problems for basic installations. The write-them-until-they-die endurance tests that have been posted here and elsewhere have demonstrated that most consumer flash drives have much more endurance than is generally assumed.
At scale, they're still kind of problems. Enterprise flash is super expensive which is why you usually see it coupled with tiering software at the SAN that uses enterprise flash as a cache for spinning platters to achieve
Re: (Score:2)
But what will they call the consumer devices? (Score:5, Funny)
You can't call them "flash drives" if it isn't flash memory, can you? We need a name that conveys the increased speed, and that maybe plays up the 3D aspect, where capacity can grow by expansion along the Z axis as well as the traditional X and Y dimensions.
I know! They can call them "Zip drives"!
Re: (Score:2)
Re:But what will they call the consumer devices? (Score:4, Funny)
They can call them Warp Drives
Re:But what will they call the consumer devices? (Score:4, Funny)
How about "cattle drives"? According to unnamed sources, we're all cows.
Moo?
Re: (Score:2)
Re: (Score:2)
You can't call them "flash drives" if it isn't flash memory, can you?
that would depend on how you define the word "flash"
is it impossible for this word to encompass new technologies?
Re: (Score:1)
Dirty branding (Score:2)
Don't they know it's not cool nowadays to be associated with gasoline?
Re: (Score:3)
Don't they know it's not cool nowadays to be associated with gasoline?
Performance studies indicate that you want the highest flash point
A few more links (Score:3)
TFA was very short on detail, so I went looking for more. Unfortunately, there seems not to be much more out there - everyone is reporting on the same short-on-detail presentation. Here's a few which seemed to me to have something to add:
kitguru has more pictures [kitguru.net]
pcworld has pictures of actual silicon [pcworld.com] (not that it has any visible detail)
digitaltrends has some interesting commentary (last two paragraphs). [digitaltrends.com]
Re: (Score:2)
After a bit more searching I find this [theplatform.net] as the most informative article so far. (I did spot a misprint: "500 times faster than DRAM" should be "500 times faster than NAND" - confirmed by following the link in that paragraph.)
Re: (Score:2)
Given the amount of time that it takes to work out the details, convince management that this is a feasible idea, make litho masks, process the wafers, and build an integrated prototype device, I would expect patent applications to be published by now, 18 months after the initial filing.
But I don't see obvious recent patent applications by Intel.
https://www.google.com/search?... [google.com]
Re: (Score:2)
That's because common practice is to crease a subsidiary and assign the patent to that entity. This is precisely to foil competitors who are keeping tabs on you.
Hmm, I work in the semiconductor industry (equipment supplier to Intel and the likes) and I have never noticed this. As far as I know, we apply for patents under our own name (at least the ones that I'm listed on as an inventor) and our main competitors do so, too (judging from the search reports that I get to see).
If this were a common practice, the names of the inventors must be listed, and it would not be so hard to search for inventor names that also appear on Intel patents. For a casual searcher on Go
Re: (Score:2)
Check Micron
ah, but now I get so many hits on inassignee:micron non-volatile memory [google.nl] even for only patent (applications) that were published since 2014 that I don't know where to start...
Probably this new technology is somewhere among those, but who knows which ones...
But how long will they last? (Score:2)
Re: (Score:2)
Is it better than current SSD/Flash?
It doesn't exist yet, so it would be hard for it to be better
Re: (Score:2, Interesting)
1000x the endurance, but since it's also 1000x faster, you can wear it out just as fast assuming you're maxing out the IO long enough, and assuming that they use similar wear leveling.
Still need RAM (Score:2)
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
With this particular technology, never... but even if it performed identically to flash, you would need a radical shift in computer architecture (both software and hardware) to support such a thing, and it would actually substantially increase complexity rather than decrease it. Memory management would become far more complex, I don't know why you think it wouldn't be needed at all.
Re: (Score:2)
With this particular technology, never... but even if it performed identically to flash, you would need a radical shift in computer architecture (both software and hardware) to support such a thing, and it would actually substantially increase complexity rather than decrease it. Memory management would become far more complex, I don't know why you think it wouldn't be needed at all.
You mean identical to RAM? It's already much faster than Flash. Why the hell would memory management become far more complex? Now you have HD/SDD - RAM - L3 - L2 - L1 - CPU Registers. If you take out RAM out of the picture, along with loading, swapping, etc. how would that make it more complicated?
Re: (Score:1)
Sorry, yes, I meant even if it was identical to RAM.
Memory management would be tougher because now you're using the same device for both long-term storage (which still requires file systems and all the usual stuff) and working RAM (which needs to be directly accessible by the CPU without any sort of thing like a filesystem). Dynamically sharing the two doesn't sound simple to me. Certainly no simpler than managing what pages get swapped in or out.
If you put a bunch of constraints on the setup, like always u
Re: (Score:3)
A solved problem, actually, besides magnetic core memory way back when. Many devices in the 80s and 90s used RAM
Re: (Score:2)
in the scenario this would just drop in as a replacement for RAM.
a (desktop)computer isn't going to run without RAM. how that ram is made up may change, of course.
however, mentioning memresistors in the same sentence implies that we will not have this in 2017 or even 2027........
This will be the death of mechanical hard drives (Score:5, Insightful)
If you look at Newegg and Amazon reviews, you'll find that perhaps the most reliable drives are 1TB in capacity and somewhat behind the cutting edge. Sure, you can get 6TB drives, but they're ticking timebombs. They have unacceptably high failure rates. As such, we're already on course for flash SSDs to overtake mechanical drives, because a 1TB SSD is approaching the price of an enterprise mechanical drive. The instant an even cheaper alternative comes out, mechanical drives are dead. They won't be cheaper by the megabyte anymore, and you can't trust them. Manufacturers COULD try to make them more reliable, but that would require more testing of individual units before shipping, which would increases costs even further. Indeed, the only reason mechanical drives are as cheap as they are is because MANUFACTURERS DO NOT TEST THEIR DRIVES. They are specifically designed so that they don't NEED to be tested. They have all kinds of failsafe mechanisms, vibration management, power management, temperature management, sector remapping, and they're over-engineered. A drive can be half broken, but you won't know because it's likely to keep working just fine. The ones that are DOA or die right away are really the worst of the lot and far more broken than you realize. The designers put all their efforts in at design time so as to cut manufacturing costs. But the end is very very near.
Re: (Score:2)
" Sure, you can get 6TB drives, but they're ticking timebombs. "
So if you try to take it on an airplane you'll get you computer confiscated by the TSA and your destination changed to Gitmo
Re: (Score:1)
Re: (Score:2)
Obviously you haven't ever had a redundant RAID cluster fail. I personally always suggest a redundant array of inexpensive redundant RAID disks. Or if you want to be EXTRA cautious, even mirroring that would give an extra layer of protection.
Re: (Score:2)
Have you ever actually worked with RAID? There is first of all the issue of all the redundancy failing before you realize it. Then poof, the next failure takes ALL of your data with it forever. There is also the issue of your redundancy failing, and while you're integrating replacement of the redundancy and the RAID is resilvering (which is a big strain on all the dr
Re: (Score:1)
Will they be better? (Score:2)
or will these ones also commit suicide when they can't be written to anymore?
Marty, it works! (Score:2)
http://hardware.slashdot.org/s... [slashdot.org]
http://hardware.slashdot.org/s... [slashdot.org]
http://hardware.slashdot.org/s... [slashdot.org]
http://hardware.slashdot.org/s... [slashdot.org]
Both DIMMs and SSDs in 2016 (Score:1)
Intel demoed the SSD version of the 3D XPoint device at IDF today.
If you listen to the keynote, they say that they'll be releasing 3D XPoint storage in both SSD and DIMM form factors next year.
The demo was relative to the Intel DC P3700 (their fastest NAND based SSD).
The DC (Data Center) P3700 is a PCI Gen3 x4 SSD. There may be higher performance SSDs out there but this one is extremely good.
http://www.intel.com/content/www/us/en/solid-state-drives/ssd-dc-p3700-spec.html
Using IOMETER, the results for the 3
Re: (Score:2)
Re: (Score:2)
It may have been the case where they decided to stop pushing traditional ssd because the 3D XPoint was so promising.
Buried Lede (Score:2)
So we can all look forward to massive lawsuits spanning at least five years before enough palms are greased and beaks wetted to release this magic to the proles.
non volatile RAM (Score:2)
This is getting to the point where you can use this as a non volatile RAM drive
Not really for consumers (Score:1)
These drives are 7 time faster than the fastest SSD, but also more expensive. That should be the other way around to be beneficial to PC users.
SSDs are currently too expensive to use as bulk storage, so normally you have a small SSD, which has the OS and other frequently accessed files and then you have a HDD which holds less used large files (i.e. video library etc.)
The HDD is an order of magnitude slower that the SSD, so this is really what is slowing down the system. This 3D XCross drive is making the p
Re: (Score:2)
You put your finger on it. For consumers, SSDs are ALREADY way more than fast enough, and with way more than enough endurance. The only room for meaningful improvement is to make something at least about as fast and long-enduring (if not more so), but CHEAPER.
Number of erase cycles? (Score:1)
Even with moders algorithms for remapping blocks, one of the points that keeps me laeving SSDs by the wayside is that they only have a limited number of erase cycles. The numbers have improved over the year, but it is still a kind of storage that has an upper limit of write accesses, making it undesirable for all those small fast writes (swap, database tables, filesystem transaction logs) where the speed would have made them really interesting. For me, the introduction of those flash drives has just replac