Intel Offers More Insight On Its 3D Memory (itworld.com) 115
itwbennett writes: When Intel and Micron Technology first announced the 3D XPoint memory in July, they promised about 1,000 times the performance of NAND flash, 1,000 times the endurance of NAND flash, and about 10 times the density of DRAM. At OpenWorld last week, Intel CEO Brian Krzanich disclosed a little more information on the new memory, which Intel will sell under the Optane brand, and did a demo on a pair of matching servers running two Oracle benchmarks. One server had Intel's P3700 NAND PCI Express SSD, which is no slouch of a drive. It can perform up to 250,000 IOPS per second. The other was a prototype Optane SSD. The Optane SSD outperformed the P3700 by 4.4 times in IOPS with 6.4 times less latency.
So which is it? (Score:5, Insightful)
1,000 times the performance, or 6 times the performance? Somebody needs to get the story right with the hyperbole.
Re: (Score:3)
I think they're figuring "a package with a gigabit can take one tenth of the space of a gigabit of dram, because stacking"
Hardware is rate limiting (Score:5, Informative)
This comparison says nothing at all about 3DXP except that it is much faster than NAND. With NAND, it is the NAND memory itself that limits the speed. With 3DXP memory, it is the PCIe connection hardware that is the slowest component and therefore rate limiting for the entire retrieval speed.
When Intel/Micron says that the 3DXP is 1000 times faster than NAND, they mean that it has only 1/1000th of the latency. You will never see that speed in an SSD drive. The speed of 3DXP will only be realized as a DIMM module in a custom designed server with all the software modifications optimized for it. 3DXP is revolutionary for in-memory applications running in server farms. And once Intel includes 3DXP on the die with the processor, nothing currently envisioned will be able to compete with it.
BTW, although Intel will have a great advantage using this technology, from what I can tell it was actually Micron that invented (or developed from an early purchased prototype) this memory. I'm still waiting for Micron to start telling us what materials were used and how this memory actually works. That will tell us what its ultimate limits are.
Re: (Score:2)
Wouldn't these be fast enough to use as RAM? Sure, maybe not for VRAM on a gaming rig or something but, say, for a server chugging away managing databases?
Re: (Score:1)
From reading this and Intel's constant announcement that this tech will be faster then RAM and there won't be any difference between RAM and HDD
Re: (Score:2)
As I mentioned in a separate reply. This may not seem important to some people but where I still the owner of my business, I'd be damned excited at the potential. Imagine, if you will, being able to page a huge data set for the purpose of deterministic modeling. I'd expect exponential speed increases from the days of spinning platters. Of course, there's much more RAM available already and I've been out of the industry for eight years... Still, it seems extremely likely to have some good results on lower-en
Re: (Score:2)
Well, 3DXP is random access so in a sense it is already RAM. It will be about 1/10th as fast as current RAM but it will be cheaper and more dense and non-volatile. So in some applications (in-memory apps in cloud servers) 3DXP may replace current RAM. An all-3DXP system might be faster than a combination of current RAM with SSD memory.
Re: (Score:2)
Thanks for explaining. I can see this as having some very valuable use cases. Imagine, to go with my earlier thought - if you will, being able to pretty much page a database that's nearing a TB in size. My background is in working with heavy data sets (traffic modeling) that there are infinite data points that one could enter. Close enough to infinite that the word is close to valid. (It's similar to attempting to model chaos.) Now, in the late 1990s, we were nearing the point where we'd be working with as
Re: (Score:2)
Well, 3DXP is random access so in a sense it is already RAM. It will be about 1/10th as fast as current RAM but it will be cheaper and more dense and non-volatile. So in some applications (in-memory apps in cloud servers) 3DXP may replace current RAM. An all-3DXP system might be faster than a combination of current RAM with SSD memory.
Write latency is the big point that Intel glosses over in these PR releases. Given that they only claim 6.4 times "latency" improvement I expect the bulk of that is write latency, measured in tens of microseconds. If that is anywhere close to the truth then it means that xpoint is definitely not RAM-like, and you won't suddenly be expecting it to save the entire state of your OS on power off because the OS would slow to glacial speeds due to orders of magnitude higher write latency.
Re: (Score:1)
They probably mean areal density.
So to answer your question, probably by stacking layers upon layers of the stuff.
Re:So which is it? (Score:4, Informative)
How is this possible? 1 bit DRAM is a single transistor.
No. A DRAM [wikipedia.org] bit is a single transistor plus a capacitor. The capacitor has to be recharged every few milliseconds, which is what makes it "dynamic".
Re:So which is it? (Score:5, Funny)
You're clearly trying to spin this to match your narrative. Well, I'm here to tell you it's a spindly argument; you can't just serve stuff like that up on a platter, step by step, and expect everyone to nod their heads. Are you tracking me solidly now? Or are you still in a state?
Re: (Score:2)
Don't be dense. We're not quite as susceptible to your magnetic personality as you believe. Now step off and voice your opinions somewhere else.
Re: (Score:2)
This is getting to be a heady argument. Why don't you two park it over there and find some way to re-float this issue? You're both coiled and it's creating a giant resistance.
Re: (Score:1)
RPM is just a fancy way to write Hz, which again is another way to express time. In the case of drives it is of interest to indicate a maximum access time.
10,000 RPM is equivalent to 6 ms and 7200 RPM is closer to 8,3 ms.
Clearly the marketing team should state that the SSD is equivalent to 2,400,000 RPM.
Re:So which is it? (Score:5, Insightful)
They claimed the technology had the potential to hit 1000 times faster than current flash memory... they didn't specify when or what flash they were comparing to.
In any case, this is an early prototype spanking the top of the line current technology. That's impressive in my book.
Re: (Score:2)
DRAM is a type of RAM.
Re: (Score:2)
Re:So which is it? (Score:5, Insightful)
Even if it is truly 1000x faster, the rest of the architecture is not designed to make full use of it.
No way any any of the current bus technologies could handle even a 10X improvement to its full extent.
Re:So which is it? (Score:5, Interesting)
There's a whole raft of other things to consider before this tech changes the IT world -- how much does it cost, how many separate fabs can produce it so there's no single-point-of-failure that could constrain supply, how much redesign of existing chipsets is required to integrate it into current server/workstation/mobile phone designs, what's the failure rate in service, power dissipation and cooling requirements etc.
Saying that the demo suggests it can be implemented into existing platforms with little difficulty. Of course as Napoleon once said, "There are lies, damned lies and rigged demos." Time will tell.
Re:So which is it? (Score:5, Informative)
If this technology can be adapted to fit into SAS-compatible packaging at MLC/3D NAND pricing this will rock the enterprise storage world for sure.
Entire brands/products in enterprise storage are built around features like caching/tiering that charge you $30k for a little flash and way more than they should for spinning rust under the promise that they'll deliver flash performance for all your workloads, most of the time.
Doing so requires beefy controllers to run elaborate tiering schemes, and along with the sky-high prices for media makes them extremely expensive and extremely profitable.
If (and this is a big if) you can get SLC durability at MLC pricing and simultaneously cut the controller cost (need less compute because you're not bothering with tiering, far less software complexity), suddenly you could have someone selling entry level 24 drive shelves with millions of IOPS and sustained transfers that will melt SAS-12 cables.
Basically it will make sense to quit using rust at all without paying nosebleed pricing at pretty much any scale.
Re:So which is it? (Score:4, Insightful)
If this tech makes it into the marketplace at reasonable prices it's not going to be hanging off SAS-12 cables or any other serial links at that rate, it will be more tightly integrated with the CPU bus to deliver on the R/W and access speed improvements. Even PCI is a possible bottleneck if this 3D flash can deliver what Intel are claiming for it. Comparing its performance to DRAM is a "tell" and shows what they're thinking; this may be the fabled "non-volatile RAM" solution that's been the Holy Grail researchers have been trying to develop pretty much ever since RAM was invented. (Yes, I know there are battery-backed-up RAM solutions that claim to be non-volatile but they're only non-volatile until the battery power runs out).
Re: (Score:2)
Early demos indicate they are thinking of doing PCIe-style drives, probably first. Which makes sense since it'll be a drop in replacement that nearly anyone can use immediately rather than as a new type of RAM that would require some pretty large architecture changes.
Re: (Score:2)
If it's cheap enough, it'd still be useful as a hard disk replacement even if it's not the most optimal way to deploy it. Fixed storage isn't going away tomorrow even if this turns out to be the holy grail of NVRAM.
I'm not sure it is, either, as its durability is compared to SSDs, not to DRAM.
Even if it was a game changer, it'd be years before hardware and upstream architectures adapted to more optimal uses for it. And if it doesn't have DRAM durability, it's more likely to be used as permanent storage an
Re: (Score:2)
This is a quibble, but non-volatile RAM has only been the Holy Grail since about 1970. Prior to that, magnetic core memory was the standard RAM technology and is non-volatile. (To quibble the quibble, for a short period of time Williams tubes were the state-of-the-art (indeed, only) RAM, and they are volatile. Alan Turing played with Williams tubes.)
Re: (Score:1)
Comparing its performance to DRAM is a "tell" and shows what they're thinking; this may be the fabled "non-volatile RAM" solution that's been the Holy Grail researchers have been trying to develop pretty much ever since RAM was invented. (Yes, I know there are battery-backed-up RAM solutions that claim to be non-volatile but they're only non-volatile until the battery power runs out).
From TFA
The company will also come out with Optane DIMMs later this year for early testers, which will combine the performance of DRAM with the capacity and cost of flash. That means a two-socket server with Optane DIMMS will have a total of 6 TB of addressable memory, "virtually eliminating paging between memory and storage, taking performance truly to a whole new level.
Seems like we're going to find out soon, 6TB of addressable non-volatile ram sounds like a game changer
Re: (Score:1)
Seems like we're going to find out soon, 6TB of addressable non-volatile ram sounds like a game changer
A server system really needs to be able to address hundreds or thousands of terabytes of storage, not just six. That's what I meant by the server system designers having to revamp the basic concepts of a computer with RAM separate from secondary storage (HDDs or SSDs on a separate bus) to one with a "flat" storage architecture. The OS will have to change too to take account of the blurring or total elimi
Re: (Score:3)
It really depends on what type of server we're talking about. Is it a front-end web server? Is it a middleware application server? Is it a database server for small to medium databases? Is it a big DB cluster? Is it a media or document storage system? Is it a hypervisor on a hardware node offering shards of its resources to VMs? These have different storage and processing needs.
In the short term, there are a few solutions for the OS and applications. Many applications will keep as much in memory as possible
Re: (Score:3)
A server system really needs to be able to address hundreds or thousands of terabytes of storage, not just six.
Only in very niche markets is that true. Most servers don't need anything near 6TB of storage, let alone 6TB of (D)RAM.
Re: (Score:2)
Most servers don't need anything near 6TB of storage, let alone 6TB of (D)RAM.
Many servers do need access to that sort of storage (and a hundred times more) and it would help if those servers run the same OS on similar hardware as other servers with less demand do. The alternative is for the sort of species differentiation that hobbles High Performance Computing (HPC) because there are few standards and a lot of hand-written system code flying in close formation, different on each machine.
I expect, if an
Re: (Score:2)
Most servers that need to access that amount of storage don't do so locally, they are doing so through things like SANs because maintaining that number of high speed low capacity drives isn't trivial and best consolidated across servers. Getting a 6TB pool would require at least 10 600GB 15k drives (the largest capacity 15k RPM drive that seagate currently makes), and that's not going into your 1x server blade, and typically isn't a great solution.
Network and remote access drives like SANs aren't going awa
Re: (Score:2)
I agree wholeheartedly that the SAN storage consolidation model isn't going away. It's logical and it's been so widely adopted with so many dollars and man-hours invested in it that it might never go away, regardless of storage device changes.
That being said, the "hyperconverged" software defined model of server nodes possessing some storage and clustering it into virtual SANs is gaining some traction. VMware has vSAN and Windows 2016 server will extend storage spaces to allow for this.
The challenge for t
Re: (Score:2)
You could do that, but probably not for the forseeable future. Files will still be needed, and a standardized way of accessing them will still be needed. Most software will likely use the same old APIs to access them, seek, etc but the driver will be seriously simple. In theory you could remove them, but then what would you do about CDs? DVDs? Backup Tapes and drives? Network connections? Network drives? Your OS will still need a unified way of accessing all that stuff, so it just makes sense to just
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
But I'm someone who believes they will deliver on some of their hype.
Re: (Score:2)
They claimed the technology had the potential to hit 1000 times faster than current flash memory... they didn't specify when or what flash they were comparing to.
Just to be up-front this is a topic I'm very ignorant about, it was only casual curiosity that made me peek into the comment section. But does't Intel normally announce a new design with where the intend for it to land long before they announce their products that are only blips along their roadmap? For example: "With this our new Pentanium Matrix we'll reach 64 cores and 10 gigahertz."... and a few months later: "Now announcing our Titticaca processor with 16 cores at 3.09ghz"
Eh I dunno, but now you hav
Re: (Score:2)
It kind of reminds me of the old "jet stream oven" infomercials.
"It is microwave fast ... 2x faster than cooking it in a conventional oven"
I'm guessing both. (Score:5, Insightful)
Increasing Memory Speeds 1000x will not lead to a straight 1000x increase in operations. There are undoubtedly other bottles necks in processing. What for instance is the theoretical max throughput of the memory interface used (is it a modified SSD interface)? What CPU overhead is involved? Don't expect your computer to perform 1000x better across the board just because one component is 1000x faster.
Re: (Score:2, Funny)
is it a modified SSD interface?
No. It'll show up using a modified DDR4 interface or an NVMe interface. You'd have to look at tech news sites (not Slashdot) to find that info.
Re: (Score:2)
Re: (Score:2)
Well you can deliver a car with 1000x the performance in horsepower, but it won't go 1000x faster. Since they've been intentionally vague about exactly what metric they based that claim on, it could be anything really. Beating a top of the line enterprise NVMe drive several times over is impressive at any rate. I look forward to seeing actual product.
Re: (Score:2)
I note that they refer to a 1000x improvement in "performance", and a 6.4x improvement in "latency". Latency is one time-related performance metric; throughput might be the other be alluded to.
Imagine two water hoses. One is ten feet long and one inch wide; the other is inches in diameter and a hundred feet long. Which can deliver water "faster"? Well, when you turn the spigot on water comes out of the ten foot first; but if you're filling up a swimming pool the hundred foot long hose is faster.
Re: (Score:2)
Neither, cause I only have a 5/8" spigot, so neither will hook up to it and I'll drag out my old hose from the garage.
Re: (Score:2)
Darn good discussion, all.
Bla bla bla (Score:2)
Look, just shut up and start shipping product. The IT community will come up with their own performance figures.
Re: (Score:2)
Yeah - at the time they even had the nerve to claim they'd be 600% accurate.
6.4 Times Less Latency (Score:3)
"6.4 times less latency" means that if the latency of the baseline thing you are comparing against is X, then the latency of the new thing has a latency of 6.4 times X less than X, which is X minus 6.4 times X, which is negative 5.4 X.
The latency we're discussing is a measurement of time (and up until Intel's amazing breakthrough it was always positive).
This means that Intel has discovered tachyons, invented a time machine, and violated causality in general. Either that, or "journalists" and marketers don't know what they're doing.
Re: (Score:1)
Re: (Score:2)
It has 6400 less millitimes as fast!
Re:6.4 Times Less Latency (Score:5, Insightful)
I think they mean that the old way takes 6.4 times as long as their newest toy.
Re: (Score:1)
Then why didn't they say that instead of the impossible and stupid mumbo-jumbo they said instead?
XPoint smells like an arch inflection point... (Score:1)
Would expect my circa 2017 laptop purchase to have 1 TB of XPoint memory, dynamically used as RAM and 'SSD' (32G and 'nearly' 1TB), deliver a 5 to 10x increase in general performance, and cost relatively the same... Reasonably stupid idea?
Re: (Score:1)
Only the "cost relatively the same" bit.
Re: (Score:2)
RAM size will decrease (Score:2)
I mentioned this in a story a few days ago, but this brings it back to the forefront. The fastest SSDs have sequential write speeds about an order of magnitude slower than typical DDR3/DDR4 SDRAM. Increasing SSD speeds to be on par with DDR means you may actually need far less RAM than you did in the past because swap operations have very little cost. If endurance ticks up three orders of magnitude (as claimed), you might start considering dropping DRAM entirely for low end computers, perhaps with an incre
Re: (Score:2)
DRAM would just become a 4th-level external CPU cache.
Re: (Score:2)
No, but you can expect your 2017/2018 laptop to have a general performance increase of maybe 10-15%, but loading things and searching for files will be 150%-500% faster depending on application, and suspend/resume may be silly fast.
Re: (Score:3)
Fix stupidity (and laziness) first.
Yeah, that's a realistic and practical precondition for any project.
Re: (Score:2)
Maybe CPU, OS and compiler tools should offer feedback to the programmer on how the program is affecting various levels of cache first, then the programmer will improve. Start with the laziness of the hardware/software vendors.
Re: (Score:2)
They do. It is called Performance counters. Check V-Tune and PAPI.
RAM Drive? (Score:2)
How does performance compare to the fastest RAM drive? And what would be the estimated cost per IOPS of each? (RAM vs SSD vs Optane)
The best is yet to come... (Score:4, Interesting)
We should get away from mass storage altogether and use this as replacement for RAM. It will take a rethinking of operating system structure, but promises to provide instant on computers with all programs and data always loaded and ready for immediate access. Database systems would immediately be orders of magnitude faster because all data is always ready for access.
I for one will not miss virtual memory...
Re: (Score:2)
We should get away from mass storage altogether and use this as replacement for RAM.
It doesn't have enough write endurance to do that. You could burn the stuff out with a FOR loop.
Re: (Score:2)
I guess we're back to VM until endurance can be addressed.
Re: (Score:2)
Re: (Score:2)
RAS syndrome (Score:1)
250,000 IOPS per second, right up there with your LCD display, PDF format, and PIN number.
Re: (Score:2)
One I've been seeing more lately that you can add to that list is "LED light."
Its The pcie interface people (Score:2)
Yes, it is only six times faster- it has probably saturated the PCIe interface. Intel has already said as such that this would be an issue and that a new interface will be needed to accommodate the rams capabilities.
Under-Hype? (Score:2)
Am I the only one who thinks this technology is shockingly under-hyped? It eliminates a 50-year old performance anchor, neutralizes the biggest challenge in Computer Science, and makes a supercomputer out of an SoC.
It came out of nowhere, but I believe Intel's claims. They wouldn't restart memory manufacturing in their own facilities if the tech wasn't ready for prime time.
Re: (Score:2)
No, I don't think so. ... about a hundred days. RAM cells get updated on the order of milliseconds rather than seconds.
At one write per second per memory cell, a device with 1000 times the endurance of NAND would last
This tech won't reduce the need for RAM. It's a better NAND, not a replacement for RAM. It may allow reduction in the quantity of RAM required where performance is not critical. It's very welcome, but no miracle cure for our computing ills.
It may also enable a renaissance in Harvard-architect
Re: (Score:2)
I see that endurance spec now. DRAM appears to be safe until they can address that.
I was imagining this would create a chip w/ 1 TB of L3. I guess that's a breakthrough or two away yet.
Re: (Score:2)
This tech won't reduce the need for RAM. It's a better NAND, not a replacement for RAM.
If it performs anywhere new hype levels then it is indeed a replacement for the RAM that would otherwise be used to cache flash or disk. For a lot of use cases, that means most of the RAM in the box.
Re: (Score:1)
Re: (Score:2)
This reminded me of the fiasco technology Bubble Memory.
You remind me of somebody who can't read numbers.