Although some industry observers – as The Reg recently noted – say that flash memory is approaching a technical brick wall, the cofounder and former CEO and chairman of SanDisk sees things differently. "Industry experts – even the best ones – are often shortsighted, and sometimes outright wrong," said Eli Harari at ISSCC on …
Meh, not really.
I think saying that NAND will replace DRAM is kind of silly. DRAM will go through advances also. And to say that ten times more NAND memory, although slower, will be better than faster DRAM memory, ah no.
I do think that just maybe flash drives will replace disk drives but that depends on where disk drives go. Today, no way, my disk drives are at least as reliable as my SSD. Sorry to say.
Re: Meh, not really.
It is sensationalist but partially true. DRAM needs seem to have plateaued on my systems and with the sole exception of my dev machine that generates large tables, none of my machines exceed 3 GB memory usage and moving my system drives from regular HD to SSD had more effect on the overall machine performance than doubling the ram from 4GB to 8GB even on my memory starved dev machine.
Not so quick
Until the SSD mfgs. can resolve chronic compatibility and reliability issues with consumer grade SSDs, they will not replace HDDs any time soon. Every major SSD hawker has had issues of one sort or another resulting in firmware updates, RMAs, lost data, reduced drive capacity, BSOD, etc. AT this point in the game consumer grade SSDs are not ready for Prime Time.
Re: Not so quick
The same has been said for the HDD industry for it's entire history and I would love for you to tell me a single HDD manufacturer who hasn't had problems. There have been firmware problems, platter problems, vibration problems, heat problems, electrical failures etc. The only reason it's news on the SSD front is because they are new.
In practice, the failure rates aren't worse than HDD and the failure points tend to happen in different places but the advice is the same as with HDD: if you care about your data then keep a backup and if a drive failure will be a huge problem for you then use RAID.
NAND replace DRAM?
I think not. Even with wear leveling, it simply won't have the lifetime to replace DRAM anytime soon.
Re: NAND replace DRAM?
What he seems to be arguing is that DRAM will essentially serve as a cache for NAND. It essentially already serves this purpose with conventional storage -- either through the abstraction of swapping virtual memory pages or explicitly loading and releasing resources on demand, but flash's superior performance could allow this "cache" to become smaller. Flash could never replace DRAM per se (for speed and endurance reasons), but I wouldn't be surprised if DRAM provisioning in PCs leveled off a bit a SSDs become more popular. Swapping is less expensive, and there is less pressure to leave commonly used resources in memory if reloading it on demand is relatively cheap anyway.
NAND shooting at the moving target of DRAM
DRAM technology keeps moving too; trying to keep up with the feeding frenzy of processors.
Computer architecture didn't make the expected leap back to "core storage" in 2008; probably because funds dried up. In 2008, it was expected that NAND would soon plug into system boards; extending NUMA space in another dimension.
The other obstacle is of course that the paradigm shift is so substantial, that the industry can't cope. Architects, hardware, systems and software are stuck up to their necks in a world where mass storage is on a different medium and needed to be fetched from cards/tape/disc/... and there's little hope that the non-technical industry participants will recognize the fundamental change; because they don't, for the most part, have an inkling about how stuff works.
Many simply can't get their minds around the fact there there is now the potential for "mass storage" to be directly addressable. It is what VM "hoped for", but now that that's deliverable at (near) full force, most don't understand how to use more than a smidgen of its potential.
We'll likely see SSDs largely replace hard disks in most PCs well before 2020. While hard disks will retain an advantage in cost per gigabyte for the foreseeable future, the vast majority of people simply don't need multiple terabytes of storage, and their storage requirements aren't likely to vastly increase in the coming years, short of some 'killer app' surfacing that requires it. Once SSDs with a few-hundred gigabytes of storage get down into a similar price bracket as lower-capacity hard drives, PC manufactures will start looking to make them the sole drive included with most systems. Certainly, there will continue to be some people storing terabytes of video and such, but that kind of storage simply isn't used in most PCs, and most of those systems will also have an SSD for a boot drive. I'm sure we'll see hard drives incorporating large amounts of flash as well. Seagate already offers some 'hybrid' drives with multi-gigabyte flash caches, and that seems likely to become the norm as disk manufacturers try to remain at least somewhat competitive with SSDs in terms of performance.
From where I'm standing SSDs have already superceded disks.
I'm sitting here with an awful box, about 4 years old. It's a pig, and was getting worse. So I bought a 128Gig Kingston SSD V+100 and asked permission to use my own drive. Then I cloned it. (It took 28 minutes.) Their techie bod, replaced in two minutes, and I was off.
I now periodically reboot just for entertainment. It takes just 35 seconds to be completely back up. Windows update takes < 1 minute. and so on.
The CPU has shot up, but it still doesn't make a 100% so who cares. It's like a new machine.
Why bother upgrading old computers when SSDs do this to them.
New computer? 1 Grand. New SSD. six score and ten.
I'm never going to buy another rotating drive.
We had that bloke from Intel last year (if not the year before) saying that by 2012 (or whenever) the majority of motherboards would have some flash soldered onto them as an extension of the memory hierarchy (registers->L1->L2->DRAM->flash->hard-disc). This would allow the DRAM that remains to be smaller and consequently faster. It seems a perfectly reasonable argument, fully in line with past historical trends.
Not buying it
I think Spin-Torque-Transfer memory will blow flash away before too long.
As fast as DRAM, long retention time, high endurance (doesn't wear out) and is inherently SLC, not the wrong-headed multi bit per cell rubbish.
I'll believe that is a good idea when Intel stops making processors using binary.
Re: Not buying it
Unless you can get large capacities for the same cost, people will go elsewhere.
The article you posted states:
"STTRAMs, and PCRAMs all have small cell size
and potential MLC capability, which could enable densities and
cost/TB comparable to that of disk drives."
Unless STTRAM is MLC the cost benefit isn't going to be there.
This is the critical factor, there are large selection of alternatives as your article states, but real world success will depend on commercialising these technologies and be able to manufacture large volumes at low costs.
Right now all of this is speculative and until you see a physical device that you can actually buy for cash out of your own pocket, assume it will never happen.
Lots of these 'ideas' come and go, the trick to success is making them stick! ;)
Re: Re: Not buying it
I think you missed how much I loathe MLC.
In my mind it only has to compete with SLC flash and it is already comparably dense with SLC NOR.
Also, it did mention STT being stackable to 4 bits per cell area, which admittedly will cost process steps.
On the other hand, it has superior power efficiency right now which is an important factor in today's devices.
"Right now all of this is speculative and until you see a physical device that you can actually buy for cash out of your own pocket, assume it will never happen." you say?
You must have missed this fine Reg' article - http://www.channelregister.co.uk/2012/01/23/dell_lsi_mram/
"...Dell and LSI are using a new type of memory in servers, RAID controllers and routers, presaging the replacement of NAND flash.
It is a form of non-volatile memory called MRAM for Magnetic RAM. MRAM has been in development for some time and is now being used for journal memory functions by Dell and LSI, according to Everspin Technologies, an MRAM developer and manufacturer.
Everspin MRAM (Toggle MRAM) is also used as journal memory on 6 and 12Gbit/s LSI RAID controller cards and included on LSI reference designs for third party RAID cards and RAID-on-Motherboard (ROMB) products.
Re: Re: Re: Not buying it
I can see why you put the eat this! :)
The reference you put the usage is as a non-volatile cache, not pure storage, wrong context.
It's replacing nvSRAM, which is a lot more expensive than conventional storage or DRAM anyway, so there is a commercial application.
My point is still valid.
Re^4: Not buying it
Well it has to start somewhere to start getting manufacturing volume up, then it can take on other roles.
A benefit of the use context replacing nvSRAM as opposed to Flash you mention, is indeed pointing out just how blazing fast it is, thanks.
And El Reg' did say "presaging the replacement of NAND flash" after all.
I have a long memory and fair patience, I may yet at some future time have the chance to say "I told you so" ;)
Anyway, sincere thanks for the debate, I miss that as a one-man embedded development company.
I'll gladly replace all my spinning disks with SSDs, but not at current prices. I can get 1.5TB of HDD for less than £100 from PC World. Ebuyer's best prices on SSD seem to be about £1/GB, so that's £1500 for the same amount of storage. Yes I would pay a premium for the better performance of SSD, but not 15 times the price.
Get the prices down and you can have a billion times the market you have today.
Best bit of all?
The limited number of write cycles means the serfs have to keep replacing stuff.
flash vs dram
where are we now, DDR3 1600 quad channel? about a combined 48GB/s? can flash match that? really?
whilst i would love an "instant on" machine, with it's memory state held in non-volatile, i don't see current flash able to maintain reliability as main memory after a run time of more than a few days. kinda like overclocking holly while asking if anyone wants any toast.
however, if we do see a merger of memory and hard disk into one all pervasive storage medium, providing instant on/off/resume and almost instant install, should this not be something to strive for?
i didn't think i'd bump against the limit after installing 6GB ram in my machine, but after installing the skyrim high res texture pack, and only running skyrim, i see my memory usage blip over 5Gb occasionally. i bet it'd preload even more if i had more ram.
"GO" for technical advancement! Go science!
The naive rush in
Some folks like being unpaid Beta testers.
I don't think I will bother listening to anybody from Sandisk due to the fact that every sandisk badged product I have owned has either been faulty or almost unusably slow.
There are other companies that I would listen to but Sandisk isn't one of them
Surely Harari knows the truth
Mr Memory quotes DDR at GBP 9 for 128 MB , flash in a USB drive costs about GBP 1 per GB . I say the cost ratio is 72 : 1 .
Are we being ripped off differentially in the ratio of 7.2 : 1
This just in...
"Technology firm's cofounder confident that his technology will prevail over the competition! Film at 11."
Maybe as a nice fast, and very much non-essential cache to the spinning disk platters, as it is used already. Maybe for small computers that don't have much storage and are meant more as fat clients. For Flash to beat HDD for me, it has to:
(a) be as reliable as a HDD
(b) be at least as cheap per unit of storage as a HDD
Neither of those are currently true, and unless all spinning-disk manufacturers conspire together to wreck their own market, HDDs are a moving target too.
Oh, and I have plenty of use for multiple terabytes or more. So does anybody that stores video.