bit-tech.net

Intel, Micron unveil ultra-fast 3D XPoint universal memory

Intel, Micron unveil ultra-fast 3D XPoint universal memory

Intel and Micron have unveiled a new technology, 3D XPoint, which promises to replace both NAND flash and DRAM in many computing applications.

Intel and Micron have announced the fruit of a joint project, 3D XPoint, which it claims can replace the NAND flash used in solid-state storage devices and offer a thousand times the performance.

Like NAND flash, 3D XPoint is a non-volatile memory component designed for permanent storage of data. Unlike NAND flash, however, Intel claims its performance is suitable for use as working memory as well - making it a potential candidate for the long-sought-after 'universal memory' that would do away with the distinction between RAM and data storage. As well as being up to a thousand times faster than the latency of NAND flash, it is only slightly slower than traditional DRAM - and has the benefit of being non-volatile and offering significantly improved density, fitting between eight and ten times the data storage capacity into the same footprint as DRAM modules.

'For decades, the industry has searched for ways to reduce the lag time between the processor and data to allow much faster analysis,' claimed Rob Crooke, senior vice president and general manager of Intel's Non-Volatile Memory Solutions Group, of the new technology. 'This new class of non-volatile memory achieves this goal and brings game-changing performance to memory and storage solutions.'

'One of the most significant hurdles in modern computing is the time it takes the processor to reach data on long-term storage,' agreed Mark Adams, president of Micron. 'This new class of non-volatile memory is a revolutionary technology that allows for quick access to enormous data sets and enables entirely new applications.'

Intel claims that 3d XPoint was 'built from the ground-up' to be a high-performance, non-volatile and high-endurance replacement for both high-capacity storage and system RAM. Intriguingly, the technology is entirely transistorless, featuring perpendicular conductors connecting 128 billion densely-packed single-bit memory cells in a lattice which is stacked two-high to form a 128Gb die and using variable-voltage selectors instead of transistors to access cells for read and write operations. Future generations, Micron has claimed, will increase the stackability as a means of boosting capacity, along with the usual lithographic process node scaling that takes place in the semiconductor industry.

Intel is clear that 3D XPoint is a real alternative to NAND flash and DRAM for big-data applications, claiming that the non-volatile memory is inexpensive and high-endurance - 'not significantly impacted by the number of write cycles it can endure,' to be precise, though no firm figures have yet been released.

While DRAM still has the edge in performance, and is unlikely to disappear for uses like processor cache and graphics memory, Intel and Micron are clearly betting heavily on 3D XPoint for the future of computing. A presentation on the technology, which is due to sample with 'select customers' later this year ahead of commercial implementations from both Intel and Micron with no firm launch date, is reproduced below.

10 Comments

Discuss in the forums Reply
Spraduke 29th July 2015, 11:58 Quote
As I understand it, this isn't as fast as D-RAM but is close enough in performance that it could be used as D-RAM if speed isn't as critical. It will be fascinating to see if this ends up in an SSD/Ram common pool of memory in certain low power applications. Would be a great way to make laptops / tablets smaller again if you just had, say, a 256GB pool of memory that was both storage and RAM.
maverik-sg1 29th July 2015, 12:00 Quote
Blurring the lines between storage and RAM is an exciting prospect - I shall be watching this one with great interest.
Gareth Halfacree 29th July 2015, 12:59 Quote
Quote:
Originally Posted by Spraduke
As I understand it, this isn't as fast as D-RAM but is close enough in performance that it could be used as D-RAM if speed isn't as critical.
Quote:
Originally Posted by The Article
As well as being up to a thousand times faster than the latency of NAND flash, it is only slightly slower than traditional DRAM [...]
:p
Yadda 29th July 2015, 13:19 Quote
Well that's funny. After ordering my (first) SSD, last week in the pub I was having an unusually constructive discussion with a friend about whether "backing storage" and "memory" would merge at some point.

Next time I see him, I'll tell him I've had a word and it's sorted. :)
schmidtbag 29th July 2015, 14:20 Quote
Wasn't ReRAM basically supposed to accomplish the same thing? What ever happened to that?


I'm a little skeptical of how this is supposed to be used. For example, let's say you open a spreadsheet that you want to edit. In today's computers, you read from the HDD/SDD and which loads the spreadsheet into memory. Any changes you make are entirely in memory, and that's the way it should be (in case you don't want to save).

With XPoint, I envision that in order to use it efficiently, you wouldn't be loading the file "into memory" because it's already there. But if that's the case, any changes you make to the file would be permanent whether you saved it or not. To avoid this, you have to emulate RAM, but that defeats using XPoint. In other words, what's the point of using XPoint as a merger of disk and RAM if it can't wholly replace them? It's slower than RAM, so the way I see it (depending your workload) you'd get better performance using XPoint for storage while using traditional RAM.

When it comes to read-only data, I think XPoint has a lot of potential. But at least in Windows, "read-only" is sometimes ambiguous.
jrs77 29th July 2015, 14:33 Quote
Call me again, when they actually release a retail-product based on this announced technology.

There's so much new and gamechanging technologies announced during the last two decades years that I've yet to see available in products that I just skip over such news these days thinking... blah blah blah.

Less talk, more actual products.
Spraduke 29th July 2015, 16:18 Quote
Quote:
Originally Posted by jrs77
Call me again, when they actually release a retail-product based on this announced technology.

There's so much new and gamechanging technologies announced during the last two decades years that I've yet to see available in products that I just skip over such news these days thinking... blah blah blah.

Less talk, more actual products.

I believe this is a product (rather than a scientific discovery) but it is being implemented into niche corners of the market just now. I imagine it would take a completely different architecture and operating system to make a true one stop shop for storage and ram.
jrs77 29th July 2015, 16:59 Quote
Quote:
Originally Posted by Spraduke
I believe this is a product (rather than a scientific discovery) but it is being implemented into niche corners of the market just now. I imagine it would take a completely different architecture and operating system to make a true one stop shop for storage and ram.

Then why not annunce a product I can actually buy? And even industry can't buy this at this time, atleast nothing in the press release from intel indicates such thing.

Additionally. If the current architecture of a PC can't deal with this new technology, then this new technology is currently worthless for me as a consumer and intel and Micron should develop a controller asap to make this new technology compatible, if they want to sell it to me.

I'm not buying into all the hype anymore until I can actually buy something.
mi1ez 30th July 2015, 00:46 Quote
Quote:
Originally Posted by schmidtbag

I'm a little skeptical of how this is supposed to be used. For example, let's say you open a spreadsheet that you want to edit. In today's computers, you read from the HDD/SDD and which loads the spreadsheet into memory. Any changes you make are entirely in memory, and that's the way it should be (in case you don't want to save).

With XPoint, I envision that in order to use it efficiently, you wouldn't be loading the file "into memory" because it's already there. But if that's the case, any changes you make to the file would be permanent whether you saved it or not. To avoid this, you have to emulate RAM, but that defeats using XPoint. In other words, what's the point of using XPoint as a merger of disk and RAM if it can't wholly replace them? It's slower than RAM, so the way I see it (depending your workload) you'd get better performance using XPoint for storage while using traditional RAM.

I see it as working with a shadow copy.
mclean007 30th July 2015, 09:04 Quote
Quote:
Originally Posted by mi1ez
Quote:
Originally Posted by schmidtbag

I'm a little skeptical of how this is supposed to be used. For example, let's say you open a spreadsheet that you want to edit. In today's computers, you read from the HDD/SDD and which loads the spreadsheet into memory. Any changes you make are entirely in memory, and that's the way it should be (in case you don't want to save).

With XPoint, I envision that in order to use it efficiently, you wouldn't be loading the file "into memory" because it's already there. But if that's the case, any changes you make to the file would be permanent whether you saved it or not. To avoid this, you have to emulate RAM, but that defeats using XPoint. In other words, what's the point of using XPoint as a merger of disk and RAM if it can't wholly replace them? It's slower than RAM, so the way I see it (depending your workload) you'd get better performance using XPoint for storage while using traditional RAM.

I see it as working with a shadow copy.
Exactly - I envisage a copy-on-write model - when you open the file in read-write mode, the OS simultaneously opens it and creates a copy in a block of free space. Any changes are done on the copy. When you "save" the file, what you are really doing is changing the pointer to the file's location on the disk, so the shadow/working version then becomes the master version. Depending what you want to do, the old version can either be retained as an old version (Apple's Time Machine is probably the most widely known example of similar tech) or released as free space. This would indeed require serious changes to the way the OS works - OSes simply aren't designed to use an amorphous block of storage as both its "disk" and its "RAM".

If this can truly be made more cheaply and with higher density than flash, I expect its first commercial deployment will be in high speed enterprise grade SSDs with custom high bandwidth interconnects; later it will trickle down to consumer products as an alternative/replacement to SSD. Don't expect it by Christmas, though ;-)
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums