Samsung Develops World's Fastest Embedded Memory With eMMC 5.0 Support

hypnosec writes "Samsung has announced the world's fastest NAND memory that supports the eMMC 5.0 standard. The new memory chips are based on 10nm class NAND flash technology and feature an interface speed of 400MB/s. Further, the 32GB and 64GB densities have a random read and write speed of 7,000 IOPS (inputs/outputs per second) while the sequential read and write speeds stand at 250MB/s and 90MB/s respectively. The chips will provide for better multitasking, HD video recording, gaming and browsing."

Comments

"The chips will provide for..."

[Anonymous Coward]

"The chips will provide for better multitasking, HD video recording, gaming and browsing."

So generic and lofty that I shat my pants. Seriously, Parity News...

Re:"The chips will provide for..."

[pipatron]

Well, to be honest, it won't provide for faster bitcoin crunching, gene folding, UFO findings, 3D rendering, slashdot posting, and other tasks popular with contemporary users of computing devices.

Re:"The chips will provide for..."

[viperidaenz]

Should it be reworded like "The chips will provide for better anything that requires the ability to read or write persistent data at any point in the process"?
That would be more accurate.

Re:"The chips will provide for..."

[Anonymous Coward]

none of the techno mumbo jumbo nerd talk. give me a car analogy instead.

Re:"The chips will provide for..."

[Anonymous Coward]

none of the techno mumbo jumbo nerd talk. give me a car analogy instead.

It's a manual, and you only know how to drive an automatic.

Re:"The chips will provide for..."

[DigiShaman]

With a standard HDD, the car's speed is slower. But most importantly, it takes time to accelerate and decelerate (the actuator arm). With an SSD, the car can accelerate instantly to 4x + times the speed limit and stop instantly as well as though you had some inertial damping field in effect. Whatever, it's FAST!

Re:"The chips will provide for..."

[Anonymous Coward]

It won't provide for faster anything I do on my computer, because I already have faster chips in my desktop. They should have just said it will make smartphones faster. For comparison, my Samsung 840 Pro SSD is rated 530 MB/s sequential reads, 390 MB/s sequential writes, 90,000 IOPS. I guess we should be impressed that smartphones and embedded devices are catching up?

Re:"The chips will provide for..."

[arth1]

Well, to be honest, it does say "embedded".

Re:"The chips will provide for..."

[gagol]

But this chip is destined to live in your upcoming cell phone, NOT your desktop. This is the news.

Re:"The chips will provide for..."

[Rockoon]

It won't provide for faster anything I do on my computer, because I already have faster chips in my desktop.

I'm pretty sure that you do not have faster flash chips in your desktop.

What you have is a faster array of flash chips, a combination that only exceeds the performance here when they operate in parallel.

Now imagine these 10nm chips in an array....

Re:"The chips will provide for..."

[Holmwood]

and "based on 10nm class NAND flash technology" is at best highly misleading. It's 19nm technology.

Parity News might better be tagged Parity Spin, as might this summary.

What Samsung is doing with NAND is actually reasonably impressive -- 19nm is very good, and their TLC stuff in the 840 looked pretty good, and the performance/reliability/value of the 840 EVO looks to be extremely good for a non-enthusiast consumer drive. Sad they feel they need ridiculous spin on top of some very respectable achievements.

Re:"The chips will provide for..."

[Anonymous Coward]

I'd rather imagine a cluster

Re:"The chips will provide for..."

[Anonymous Coward]

What kind of cluster?

Re:"The chips will provide for..."

[Anonymous Coward]

Good to see people in this thread being so honest.

Re:"The chips will provide for..."

[Anonymous Coward]

Now imagine a beowulf cluster of these 10nm chips arrays.

Re:"The chips will provide for..."

[Anne Thwacks]

Please can someone move this post to the top. Its the only useful one in the whole thread!

Re:"The chips will provide for..."

[unixisc]

Embedded memory does not mean memory for embedded applications. It means memory that is included as a part of a larger subsystem, sometimes in a multi-chip module (MCM), sometimes in a package-on-package (POP), and the 2 main reasons for it is typically real estate constraints, as well as performance. For instance, this chip is a NAND flash that could go into MCMs that include application processors or basebands in cellphones, or it could be a part of multi-memory packages, where it's combined w/ DDR3 DRAM to provide all the memory that a portable app - such as a cell phone or GPS unit - may need.

I am curious about their 'random read and write' claims - NAND flash does not do random reads or writes: it reads or writes in pages, and so an entire buffer has to be filled before one can write anything. NOR flash is what has the random read and write: one erases in sectors/blocks and programs in bits/bytes/words. Samsung happens to make both, last I checked, so it's not inconceivable that they've combined the 2, and are offering the combination in a single package. But I'd like to know whether that's the case here.

Re:"The chips will provide for..."

[K. S. Kyosuke]

A...Grendel cluster?

Re:"The chips will provide for..."

[AmiMoJo]

Small correction: NAND can read randomly. Often the controller only supports block level reads, but the memory itself does not mandate it. SD cards support byte level addressing, for example, at least up to 2GB.

Re:"The chips will provide for..."

[jones_supa]

Or another one looking at the seek time and fragmentation:

The HDD is a car with acceptable top speed, but is slow taking turns. An SSD is a car with much higher top speed and can also make instantaneous sharp turns.

Yay slashvertisement

[Anonymous Coward]

And how many write cycles? HOW MANY CYCLES?

Re:Yay slashvertisement

[Dunbal]

At least 1?

Re:Yay slashvertisement

[Anonymous Coward]

One bit at a time. Savor each and every one.

Re:Yay slashvertisement

[fuzzyfuzzyfungus]

At least 1?

With new pre-loaded Premium Content from our Exclusive Content Partners, tedious "writing" is a thing of the past!

Re:Yay slashvertisement

[arth1]

Yes, that would be Useful Information.

And also, what's the erase time?
The main problem I have with modern NAND is the worst case latency, when a sector has to be erased - a 1 second hiccup isn't good, even if it only happens rarely.
In short, I don't care about best times or average times, but want the worst time to go down, not up.

Re:Yay slashvertisement

[Anonymous Coward]

In soviet russia, write cycle wrong cycle

Re:Yay slashvertisement

[Anonymous Coward]

You honestly have to wonder if dice board members recently bought up a bunch of Samsung stock. They've been pimping Samsung out hard in stories the past few weeks.

Re:Yay slashvertisement

[Anonymous Coward]

Pff, flash memories have tolerated more write cycles than spinning platter since the day SSDs became mainstream. The problem is that the memory manufacturers didn't have the experience of wear leveling that spinning disk manufacturers had. (Remember when disks were in the 100MB range and didn't have wear leveling. Dead blocks and lost data all the time.)
Stating the number of write cycles is pointless unless you specify the wear leveling algorithm. The flash memory manufacturers should have taken a hint from the spinning disks and never specified it in the first place.

Re:Yay slashvertisement

[AmiMoJo]

Knowing the number of re-write cycles alone isn't useful. You need to know how much spare capacity is available too, and if the controller does things like compression and caching.

Re:Yay slashvertisement

[unixisc]

Typically, any flash memory has 10k cycles per sector/block. Since you have a controller, it could do something like re-map writes to unused portions of the flash, so that sectors are uniformly written, as much as possible. So depending on how that works and how well they do it, one could multiply that with the number of blocks in the flash, and get an enhanced number for the total write cycles

Re:Yay slashvertisement

[unixisc]

Erase times are always bad in flash - ranging in the order of seconds. Even in NOR flash it's bad. The only exceptions I've seen are Atmel's Di-NOR based flash - where they use 2 transistors per cell, which obviously reduces the capacity of the memory in the same lithography, and SST's SuperFlash, which is now a part of Microchip, and its original founders now running a company called Greenliant. But none of these products are NAND flash.

The reason the worst times go up is die shrinks - in order to avoid overcharging or damaging the cells, the charge pumps are made more sensitive and use more precharge cycles to set off the erase mechanism. There is no getting around shrinks, since the flash manufacturers are always under pricing pressure, and only enjoy a respite when there is a market shortage. So that's not a problem you're gonna see go away.

Re:Yay slashvertisement

[ultrasawblade]

Citation needed. What spinning HDD uses wear leveling? Unless you mean sector sparing, but that's something different.

Re:Yay slashvertisement

[Lunix Nutcase]

HDDs do not use wear leveling. At least not "wear leveling" as applies to SSDs and Flash drives.

hmm

[Trepidity]

These numbers may not make sense to everyone and if we translate them to ‘English’, Samsung means that these chips will provide better multitasking, browsing, HD video recording, gaming, file transfers – all in all a performance boost as compared to today’s chips.

A better translation could be to give me some information about what the current marketplace looks like. If it's the "fastest embedded memory", is that because it's 20% faster than the existing parts? 2% faster?

Re:hmm

[Anonymous Coward]

Well, let's say this Twinkie represents the normal number of I/O operations per second in the mobile arena. Based on this article's sample, it would be a Twinkie... thirty-five feet long, weighing approximately six hundred pounds.

Re:hmm

[Anonymous Coward]

It is about 0% faster for reads than just-released products, while about -50% faster for writes and -70% faster for IOPS.

For reference:

"Featuring 240GB of capacity, the RAIDR Express attains sequential reads of 830 MB/s and writes of 810 MB/s. ASUS ROG rates the RAIDR Express at 620,000 hours mean time between failures (MTBF), which equates to incredible durability. Random 4K read/write IOPS are stated as 100,000. The RAIDR Express utilizes dual LSI SandForce controllers (latest generation) and 19nm Toshiba multi-level cell (MLC) synchronous NAND Flash."

This is comparing manufacturer-claimed numbers to eachother, so the numbers WILL be high, but the ratios should be about right. I'm assuming 64GB chips in use, so 4 of them get you to equivalent space, and I'm assuming 100% scaling. Using 32GB chips at the same speed means 2x as many chips, so 2x the speed (at 100% scaling).

Re:hmm

[Rockoon]

It is about 0% faster for reads than just-released products, while about -50% faster for writes and -70% faster for IOPS.

That doesnt seem to be true. Those produces use many chips to attain their (essentially they are a RAID-0 of many flash chips) , while this is a single chip.

Re:hmm

[Anonymous Coward]

THAT'S A BIG TWINKIE!

Re:hmm

[Anonymous Coward]

A single chip ... containing 8 flash die + controller.

Re:hmm

[Anonymous Coward]

Not only do those products use multiple flash chips as a more or less RAID-0, they also include DRAM to cache transactions. Especially the write IOPS are not comparable, since 32 write transactions can be fused into a single 128k write to a single Flash chip. That alone is an order of magnitude larger than your -70% IOPS claim.

Re:hmm

[Anonymous Coward]

Re-read my post. I multiplied all the quoted numbers by 4, because you'd have 4 of these chips to get 256GB of space, approximately equivalent to the ASUS product compared against. You can make assumptions that affect the numbers however you'd like, but I did list the ones I was making, which I felt were friendly to the new Samsung product (e.g. 100% efficient performance scaling).

Multimedia Experience

[Reliable Windmill]

Will they also provide a richer multimedia experience, more vibrant colors, and increased productivity? I hate these dumbed-down explanations of the benefits of some new computer technology.

Re: Multimedia Experience

[LodCrappo]

Everything you do will be faster and more fun.

Re:Multimedia Experience

[jones_supa]

Aagh, this. Makes me want to jump out of window when I see those laptop reviews where the screen is described to have "vivid colors".

All I want to know is

[adolf]

How well does it fellate?

Re:All I want to know is

[reboot246]

I would guess . . . way too fast?
You'd probably get a really bad burn.

Re:All I want to know is

[Megahard]

Just be sure to contact your doctor if you have an erection lasting more than four hours.

Another Slasdot paid ad

[Anonymous Coward]

QUOTE "The chips will provide for better multitasking, HD video recording, gaming and browsing"
Would anyone like to make an attempt to justify any of the above claims?

-Multitasking? On a modern machine, this will be a RAM and CPU issue.

-HD video recording? Yes, if you are Peter Jackson working on the next 'Hobbit' movie. For every ordinary users of HD video cameras, the camera pre-compresses the data stream to a level well below the memory bandwidth of existing high-end flash cards.

-Gaming? SSD certainly do improve some gaming experience to a limited extent- usually the speed with which the next level is loaded in a multi-player FPS. However, the increasing use of SANE programming methods like streaming make such high-speed memory blocks almost redundant. This fact, and the increasing amounts of memory found on the GPU card itself mean that the famous SSD gaming boost has really peaked.

-Browsing? Again a RAM and CPU issue.

Faster is nice if it costs no more, otherwise we turn to computer science theory to see if the additional memory bandwidth at that stage of the computer pipeline really can make much of a difference. In this case, unless you are constantly accessing large new datasets (which might be the case with high-definition video EDITING, or many server database situations) the boost in flash speed is essentially of little significance.

Re:Another Slasdot paid ad

[fuzzyfuzzyfungus]

I agree with your general sentiment, that the marketing puff piece is a marketing puff piece, and larded with nonsense; but it is worth noting that this is an eMMC part, not a chip destined to do the behind-the-scenes work in a SATA or PCIe SSD (though Samsung presumably has a design that exploits the same flash cells with a different interface either available or in the pipeline); which very strongly suggests that it's aimed at mobile devices.

Compared to proper computers, even the latest mobiles tend to be pretty stingy on the RAM (both in terms of how much gets baked onto the board, and in terms of how aggressively mobile OSes take apparently-idle tasks out back and shoot them), so 'multitasking' can, plausibly, mean a workload that involves lots of re-loading applications from nonvolatile storage because the OS killed them.

It doesn't make the PR puff any less puffy; but expected eMMC use cases can't reasonably assume that 16GB of RAM is cheap as dirt and 64 bit addressing is a feature you can't even avoid buying; because neither is remotely true, poor bastards.

Re:Another Slasdot paid ad

[Rockoon]

SATA SSD's dont need faster flash at this time because the SATA consortium was too short sighted to see the need for a much fatter pipe, and because thats the bulk of the desktop SSD market, nobody is developing faster flash intended for desktop use. Mobil devices on the other hand...

This will eventually bleed over into PCIe solutions, but its hard to imagine extensive assembly lines for the purpose.

Re:Another Slasdot paid ad

[bill_mcgonigle]

This will eventually bleed over into PCIe solutions, but its hard to imagine extensive assembly lines for the purpose.

Do we have server boards with flash slots alongside the RAM slots yet? That has to be coming - I keep hanging SSD's, which are mostly packaging, off of SATA cables for caching purposes and it seems silly.

Re:Another Slasdot paid ad

[SuricouRaven]

Multitasking: Swap. Notice that most applications on smartphones cease to execute when not actually on screen? The OS puts them onto flash to free up precious RAM. Not much RAM in a phone.

HD Video recording: But now it doesn't need to compress it so much. Quality can improve.

Gaming: Er, not so much. Faster level load times, but that's about it.

Browsing: Now they are mostly making things up. I suppose it could speed up cache access, but that's hardly a bottleneck.

Re:Another Slasdot paid ad

[fuzzyfuzzyfungus]

I don't think that anybody has defined a special, application-specific, 'flash slot'; but pretty much all the non SATA/SAS drives (that you'd see in a server, things like this eMMC chip not so much) are just PCIe cards, and those are common enough, and often not otherwise occupied.

It is true, though, that servers specifically built around the mechanical requirements of shoving a bunch of PCIe cards in are markedly less common than ones build around the mechanical requirements of shoving a bunch of HDDs in (and I don't think I've ever seen a server system built around the notion of using miniPCIe SSDs; that would be brutally expensive; but those things are only about the size of a DIMM, so even a 1U could accommodate pretty alarming capacity without using a proprietary form factor or socket...)

Re:Another Slasdot paid ad

[AmiMoJo]

To be fair SATA was developed at a time when faster speeds and smaller cables were required, but backwards compatibility and low cost were also primary considerations. PCI-e is a fundamentally different technology, designed mostly for throughput over short distances in an electrically controlled environment.

In SATA's case simply increasing the bandwidth available is only half the problem. There is overhead and lots of signalling, with asynchronous operation of commands and the need for data integrity checks.

Re:Another Slasdot paid ad

[dfghjk]

The "SATA consortium" was far-sighted enough to know that pushing beyond SATA-3 speeds would simply be duplication of effort. The desktop market is not suffering from single pipes limited by 6Gb speeds and the eventual successor will be PCIe-based. The future of desktop storage is not obligated to be SATA and they know that. Apparently you don't.

SAS is doing a 12Gb PHY, so if SATA turns out to want it, which they won't, it will be there to take. SATA3 will be fine for hard drives, SSD is better off on Express.

Re:Another Slasdot paid ad

[Anonymous Coward]

eMMS is used primarely in phones. Note for example that the current iPad (4) has a peak speed of an astounding 40 (forty) megabytes per second.

Re:Another Slasdot paid ad

[tlhIngan]

To be fair SATA was developed at a time when faster speeds and smaller cables were required, but backwards compatibility and low cost were also primary considerations. PCI-e is a fundamentally different technology, designed mostly for throughput over short distances in an electrically controlled environment.

As was PCIe - it was designed in an environment where PCI was limiting - most PCI implementations were stuck with 32-bit 33MHz, despite enhancements to 100MHz (33/66/100), 64-bit, and other things. The problem was it wasn't scalable, and they needed a backwards-compatible mechanism for the next-generation of interconnect busses. (PCIe is logically compatible with PCI for legacy OSes)

In fact, PCIe was designed to be more forgiving than PCI, like how SATA is.

Multitasking: Swap. Notice that most applications on smartphones cease to execute when not actually on screen? The OS puts them onto flash to free up precious RAM. Not much RAM in a phone.

There's plenty of RAM in a phone - 1GB, 2GB are common now in the high end ones.

The reason mobile OSes suspend apps in the background is battery life - an app running in the background can easily gobble up enough CPU to drain the battery pretty darned quickly.

And most mobile developers are not power-aware enough to write good battery sparing code. It's far easier to suspend an app (even in RAM) than to keep it running. Those that need an app to run in the background can use the OS APIs to declare the intent to do so.

To not do so makes for a rather pathetic user experience - if they do nothing, their phone lasts all day. If they run some apps, suddnely they get 3 hours of battery life because one of those apps went wild. And they aren't going to investigate the issue - all they know is they have a crap phone where randomly one day you'll get great battery life and the next it'll die before noon.

Re:Another Slasdot paid ad

[bill_mcgonigle]

but pretty much all the non SATA/SAS drives (that you'd see in a server, things like this eMMC chip not so much) are just PCIe cards

Right, and if you look at a 1U machine they have one or two of them on a riser card, if any. The same machine might have 8 2.5" bays in the front of it.

the notion of using miniPCIe SSDs; that would be brutally expensive; but those things are only about the size of a DIMM, so even a 1U could accommodate pretty alarming capacity without using a proprietary form factor or socket...)

That's at least a good start! Other than mass-market effects, is there any reason for a MiniPCIe part to be more expensive than an SSD-form factor part?

Re:Another Slasdot paid ad

[fuzzyfuzzyfungus]

I imagine that the comparatively strict board size limit forces you to use the densest NAND packages to achieve reasonable capacities, and also limits the number of independent NAND chips you can have running in parallel behind your controller chip, so you may have to use faster NAND than some of the physically larger drives.

Aside from that, and maybe a couple of extra PCB layers, I think that it's mostly a question of volume.

How an SSD could speed up 3D rendering

[tepples]

I agree with most of your examples, but I can think of a situation where an SSD might help with faster 3D rendering. The soft-real-time 3D renderer in a video game is often bottlenecked by the speed of loading textures into RAM. If you've ever seen the blurfest that is the start of an Unreal Engine 3 level before the textures pop into focus, you know what I'm talking about. There's a reason that PC games load faster when installed to SSD.

Re:How an SSD could speed up 3D rendering

[gagol]

When we talk about videogame, we talk about gaming. This is about pro 3D modelling and rendering it to video. Not everything is about you in this universe.

Re:How an SSD could speed up 3D rendering

[InvalidError]

Even better: put enough RAM in your PC so all the files get loaded into the OS' disk cache. This won't help the initial load time but once everything is loaded into disk cache, applications and games hardly ever have to touch the HDD/SSD again until you reboot your PC or load some other big game or application.

When gaming sites do computer reviews and evaluate the impact of extra RAM on games, they should benchmark the initial load time separately from reloading times. I bet more RAM would have nice performance benefits even on SSDs in such a scenario.

SSDs may be faster but not as fast as having enough RAM that you can bypass SSD accesses altogether.

Re:How an SSD could speed up 3D rendering

[Anonymous Coward]

Okay, how do you put more than 32GB on consumer hardware?
And no, $1k rebadged E5 xeons don't count.

Re:How an SSD could speed up 3D rendering

[TapeCutter]

I can think of a situation where an SSD might help with faster 3D rendering.

Yep, an SSD on my i5 makes WoT play at "highest detail" just as good as it does on my i7 using a conventional HDD, they both have the same video card and ram but without the SSD the i5 is practically unplayable (especially if you want all the eye-candy). It also loads the O/S and game faster than the i7. However durability is a bit of a concern, my first SSD shit itself without warning after 3 months, it was replaced under warranty and has been running for about a year now without problems. "Windoze" gets a lot of bad press but I have to say the is a remarkably simple and useful tool for determining the best way to spend your hardware upgrade money.

Re:How an SSD could speed up 3D rendering

[TapeCutter]

Who pissed in your cornflakes? The underlying reason it speeds up both games and rendering is identical, ie: swap space.

Re:How an SSD could speed up 3D rendering

[K. S. Kyosuke]

This is about pro 3D modelling and rendering it to video.

And, actually, one of the design considerations in developing the REYES algorithm for Pixar PRMan was storage coherence. In fact, the whole thing was designed around the need for data access coherence.

Re:How an SSD could speed up 3D rendering

[jones_supa]

DragonFlyBSD has an excellent feature which allows you to use a SSD as filesystem cache, so you get a secondary level FS cache after the RAM FS cache. Then you could have for example 4GB RAM, 32GB SSD and 1000GB HDD.

Re:How an SSD could speed up 3D rendering

[tepples]

Real-time rendering to video, as might be seen in a live television broadcast, has similar constraints to video games.

Re:How an SSD could speed up 3D rendering

[Anonymous Coward]

First: you will be able to soon, there are 4 slot boards that have maximum memory at 64GB (there just aren't 16GB ram sticks widely available yet). Secondly: what game cannot be cached in 28GB of ram? (leaving another 4GB for things like the OS)

Re:How an SSD could speed up 3D rendering

[Anonymous Coward]

Posting AC as I can't be bothered logging in. You can load as much data as you want in to RAM and have amazing algorithms predicting what the CPU might want next, you still can't guarantee finding the data the processor is looking for. Understand that 90% of the time the piece of data the CPU wants is in L1 cache. If not, 90% chance it's in L2. If not, 90% chance it's in L3. If not, 90% chance it's in RAM. That sounds great, until considering how many operations a CPU does in a second. The search time dramatically increases once you leave RAM and is the biggest bottleneck in modern computer systems. It needs to be tackled.

Re:How an SSD could speed up 3D rendering

[InvalidError]

Or forgo the 32GB SSD and have 32GB RAM instead.

I have 32GB RAM in my current PC - I always end up maxing out RAM on my PCs so I decided to bite the bullet while DDR3 was still near its all-time low.

I paid $500 to put 512MB in my P3 13 years ago - $300 for the first 256MB, $200 for the next 256MB a year later.
I paid $400 to put 3GB RAM in my P4 9 years ago - $250 for the first 1GB, $150 for the next 2GB two years later.
I paid $325 to put 8GB RAM in my C2D 5 years ago - $200 for the first 4GB, $125 for the next 4GB two years later.
I paid $185 to put 32GB RAM in my current i5 - $90 for the first 16GB, $95 for the next 16GB a few months later.

Disk IO has always bugged the heck out of me so I have always built my PCs around eliminating as much of it as I can. I'm quite happy with how my current PC hardly ever needs to access the HDD once all my programs and their data are loaded.

As long as...

[Anonymous Coward]

... they fix the eMMC sudden death problem.

Re : So what? Tell us something pathbreaking.

[The Aditya Nath Jha]

Such innovations are being made daily. It did not redefine computing nor is a leap signiifcant enough to close my other 12 tabs just to see this story :D I think I should stop giving Slashdot so much attention :P -- And I thought my jokes were bad!

Sandy Bridge.

[Anonymous Coward]

4x8GB == 32GB.

Re:Sandy Bridge.

[omnichad]

== != >. Filter error - explanation - == does not mean greater than.

Flash slots?

[unixisc]

Honestly, why do we need a separate flash slot? It sounds like when Intel introduced the AGP slot, which is no longer there on motherboards today. The data transfer rates of PCIe is adequate for getting that data to the flash, which will in any case need some buffer chips in b/w, since there is no way NAND flash can such up data at PCIe rates. There have been, from what I understand, market research done into whether NOR flash should have a DDR like interface as DRAM does, but that's for applications like smart phones, not for memory modules, which are supposed to be as cheap, if not cheaper, than DRAM.

Essentially, the solution here - have a controller chip that translates the PCIe protocols to the ones that the NAND will understand, in addition to ECC and all the rest. Then the cost of that chip would determine how much of NAND can be stacked on that card and still be cost-effective.

Re:Flash slots?

[bill_mcgonigle]

since there is no way NAND flash can such up data at PCIe rates

That's the right metric - I was under the impression that some of the fancier NAND arrays (FusionIO and the like) were already limited by PCIe performance and could use faster access to the bridge for DMA purposes.

It could be that I heard that story about PCIe2.0 though.