Intel/Micron 3D Xpoint Memory

See:
http://www.tomshardware.com/news/micron-intel-3d-xpoint-memory,29690.html

Intel just keeps impressing me.

If they can deliver this for retail purchase and make it compatible and reliable with existing platforms, wow.

If they have to make it work with Skylake and forward platforms - still hugely impressed.

Now, give us 1 Exabyte of this tech and stand back world!

Low (much closer to cpu-like) latencies. Increased 2 magnitudes increase in endurance and non-volatile too.

This, at the right capacity, is worth $$$$ to me right now - sight unseen.

Who else is doubting Intel's dedication to providing users with the best tech possible today?

They may have even delayed Skylake for this tech to have matured sufficiently (internally) so that the new platform can really take off.

I wonder what other tricks they have up their sleeves in the next little while?

(Btw, I also like the stats quoted below):

I can't wait for 2016 to get here. It seems like I'm trying to build a rocket ship with steam engine parts today.

 

See:
http://forum.notebookreview.com/thr...point-cross-point-non-volatile-memory.779212/


The link found by alexhawker has even more details (thank you!).

When I first read the story, I couldn't believe that Intel had kept such a huge advancement a secret. I really was prepared to admit it may have been a hoax. Glad I took a chance and posted it though - .

The BBC link puts this new tech into perspective.

And that is...

Baby steps. yeah.

DRAM is still faster than X-RAM(TM), even if X-RAM is faster than any of today's SSD's in any form.

That makes it the perfect X level cache to introduce into tomorrow's systems - right between the RAM and the SSD. Wonder if MS is staying up at nights to implement it in this way?

I have to admit; I'm finally not bored of the tech news anymore. Each new bit of information on this new tech/component will be thirstily inhaled and digested/analyzed thoroughly.

Will OP'ing not be needed anymore? Will OP'ing be even possible? (To both; doesn't seem likely).

Will a PerfectDisk/MyDefrag run on the X-RAM make a difference as it does now with any SSD I have tried it with? (I want to play with some now...).

If the capacities introduced start at the 512GB level or higher; all the NUC's I'm using myself and helped sell/setup to my clients will be instantly upgraded.

I wonder if this persistence indicates the potential to forget about UPS devices in certain setups?

The very low power (that should be) needed for this 'simple' device should really kick SSD's in the pants with their HDD-like power requirements (at any load above idle). This, along with improvements on Skylake's successor may finally enable a true digital notebook (and/or phone... finally) that can last a week or more in real world (and actual) use.

Looking forward to seeing more info trickle out about this.

Nobody has any thoughts on this except for me and alexhawker?

 

tilleroftheearth, they won't introduce it as long as older tech sells well.

 

Now it makes more sense why they put eDRAM on package. All the deal about "graphics" didn't seem worth it for the R&D and silicon used for the eDRAM. The real reason must have been so they can use eDRAM as a cache for the Xpoint memory and basically have the benefits of DRAM and persistence of NAND/Hard Drive.

Boot: Instant, because its on the Xpoint memory, no worrying about sleep modes, and power drain, no such thing as boot problems.

When booted, certain performance sensitive code caches into eDRAM, and it can be used for both CPU and Graphics

The exact scenario I am talking about was mentioned by an Intel investor at an investor website. Also Purley server platform is said to have "persistent memory". From what it looks like it uses regular DDR4 DRAM memory as some sort of a caching agent along with probably Xpoint memory.

Now this seems worth the upgrade. Perhaps Kabylake, or Cannonlake for PC users?

 

You seem to be wrong already.

They are gearing up/producing it as we speak and some form of it will be ready/shipping in 2016.

 

I don't see how this is related to eDRAM? It has no persistent storage capability and therefore no storage ability at all?

Putting the RAM on the cpu die may be just a step towards the real goal (gen 2,3 or 4 of X-RAM). But I doubt that X-RAM will affect the igpu to any degree - that is what the L4 cache is for.

And the L4 cache affects the igpu immensely.

 

I wonder if the underlying technology is RRAM. The diagrams are eerily similar to those put out by Crossbar's RRAM announcement half a year ago. They mention changing resistence. Even the name is similar. CrossPoint <==> Crossbar.

Wikipedia article on RRAM: https://en.wikipedia.org/wiki/Resistive_random-access_memory

As much as I root for the startups, it's good to see a big player take the post-NAND era seriously.

 

No, the reason is because Xpoint is not up to the speeds of DRAM. So to be a DRAM replacement for high performance solutions, it needs some caching with DRAM. It's fast enough that its viable, but not enough alone.

Why I mentioned about eDRAM: Currently eDRAM is a very low volume solution and likely not impacts their bottom line much if at all to any meaningful amounts. They do not put so much efforts just for so little market, that is high performance iGPU. It's extremely silly. There had to be a bigger reason for that, and this is it. Graphics is merely a cover. But this will benefit more than just graphics.

The patents Intel has talks about this. Their server platform Purley hints at this. Semiaccurate has talked about it: http://semiaccurate.com/2015/05/20/another-intel-memory-code-name-pops-up/

The link from the article points to yet another article titled: "Intel to do away with DRAM in PCs".

But this won't happen right away, because the tech won't be fast enough. Not yet.

 

The good news is that sample are coming to "select customers" later this year so this is almost certainly quite real and not something that is hyped for years without ever being seen in a commercial product (e.g. memristors). The bad news is that so far all usage mentioned has been in server and HPC environments so it is also almost certainly quite expensive and will not be seen in consumer products for quite some time. Still, it seems to scale so if it does what they say, it will come to us eventually.

 

I wouldn't expect to see this in Skylake except in some form of cache. It's too young to be ready for consumer level use. But if it can be produced in volume and cheap enough SSD's as we know it are history, and pricing should drop considerably as well. Now we should see easily see 2TB+ super fast storage for mere pennies per GB in the next couple of years.

 

I was hoping it could appear in the Batman 2.0 HTWingNut. Btw do you know whether the 750/770ZM can be NVMe enabled via BIOS? My understanding is that it will be standard PCIe host protocol for Xpoint. Is it likely that Clevo will be generous or do we have to count on Prema's magic?

 

Too young? Nah. A decade of Intel R&D and persistence is enough for consumer use.

 

X-RAM is not up to the speeds of the best DRAM, but it is more than fast enough - especially when combined with persistence and the fastest (at least for latency) storage yet.

A high performance (or high enough...) igpu is not a small market. Especially when they can take a 100W+ discrete component and include it on the cpu die for a savings of 40% power usage... - including the CPU... No, we're not quite there yet, but a few iterations more and we'll have surpassed that goal.

On die RAM is already a reality (well, almost... ).

See:
http://www.digitaltrends.com/computing/new-details-revealed-about-intels-upcoming-60-core-behemoth/

At up to 16GB and 400GB/s speeds (many times over what DRAM based RAM solutions today offer).

RAM is not silly. It is an equal partner to the CPU (one without the other is useless and you don't have a computer, you have a paperweight).

On die fast RAM for the igpu is not silly either. It enables efficiencies in power usage while increasing the performance too.

If DRAM could be 'done away with' - it would have been. Even if it was marginally slower. That would take huge $$ off each system produced which effectively and easily increases the margins significantly and blatantly.

All computer systems need a CPU plus RAM to get any meaningful work done. This will not change anytime soon with the current and known/near/future platforms.

Your confusing X-RAM with the L4 cache used in Iris Pro graphics is making me confused?

A fast(er) enough cache (L1/L2/L3/L4/L'x'...) will always be useful to the almost too fast cpu...

But thinking that Cross Point Memory is the final destination of the Iris Pro RAM is more than a little far fetched at this stage of the game (I can almost see this in ~50 years from now, but technology will change the computing platform (including the cpu and ram interaction as we know it) more in that time frame than it will keep the same).

 

There's been some interesting speculation as to what the underlying technology is. The consensus seems to be it's either a form of PCM or RRAM.
http://thememoryguy.com/micronintel-3d-xpoint-raises-more-questions-than-answers/
http://www.eetimes.com/document.asp?doc_id=1327289
http://www.dailytech.com/Exclusive+...emory+Boy+Did+They+Patent+It/article37451.htm

As the dailytech article mentioned, this seems like it's connected to Purley:
http://wccftech.com/massive-intel-xeon-e5-xeon-e7-skylake-purley-biggest-advancement-nehalem/
This would fit in well with the " All new memory architecture" that's " up to 4x capacity & lower cost than DRAM" and " 500X faster than NAND."

If true then the first products will probably be part high end Skylake-E servers in about a year. Timing-wise that'd make sense.

As for consumer products I wouldn't hold my breath. This would neither be replacing DRAM nor NAND, so it'd be creating another storage tier inbetween. There'd have to be some massive changes to OS kernels to be able to take advantage of that.

 

Thanks superparamegnetic for the additional info and links, but I think I remember reading specifically that it was not RRAM or anything else except a brand new tech?

 

Based on that info, this stuff is far from ready for prime time. It still needs to stretch it's legs for a few years before it makes its way into consumer application.

And if it's 1000 times faster than NAND wouldn't that make it faster than DRAM? I mean from a practical perspective a PCIe M.2 drive can run 1000MB/sec. This means this would be 1000000 MB/sec (one million MB/sec or 1TB/sec). I don't know of any RAM that is that fast.

 

http://www.anandtech.com/show/9470/...000x-higher-performance-endurance-than-nand/5

Far from prime time? I think not. It is in production and volume ships next year. We're just still in the dark of what product(s) they'll actually ship (Intel is playing this very close to their chest...).

1000x faster is with regards to latency . X-RAM is 10's of nanoseconds. Nand is 100's of microseconds.




The key point is that it is much closer to DRAM than NAND, with regards to latency.

DRAM has effectively unlimited endurance and this is what X-RAM needs to match, if it can, in the coming years.

But combining the best of DRAM and NAND and having it (eventually) available as a PCIe add in card with NVMe support (Intel's most likely initial offering, but still somewhat performance limited...) and as a memory like DDR4 directly connected to the cpu offering (Micron...), this will change many aspects of computing in the years to come. Including software algorithms, games/consoles and future paths of PCIe lanes a cpu can/will include.

This puts the expectations of AMD and HBM into perspective: yawn.

 

While XPoint is currently being marketed as a "go between" for DRAM and NAND, I can definitely see it as a strong contender for a NAND replacement once Intel & Micron get capacity and production ramped up. And I agree with Anandtech's perspective: the fact that this is coming from established manufacturers is very good news in terms of getting it to market and seeing it in products.

 

The AnandTech article with all of the slides clarifies a few things. It's certainly not for consumer products in the short term and it's not being positioned as a replacement for NAND (which presumably means that the cost is too high). Still, it's a potentially useful product: my workplace could have used it a couple of years ago when we had a scenario that would almost have best been served by SSDs, but there weren't any with the endurance that we needed so we wound up falling back on ramdisks. Also, if they can get the manufacturing issues worked out, it might come to consumer devices in the longer term.

 

Just what we need. Intel to have a stranglehold on the RAM/NAND segment as well. Boo. I hope Samsung or other silicon giant respond with something more significant and cheaper in the next year.

 

Wishing Samsung to do something worthwhile for their end customers is wishful thinking, but either way, Intel doesn't have a stranglehold on this tech - Micron is what is currently keeping the tables balanced.