isuppli: Will $1 per GB NAND Resuscitate Solid State Drives?

[cbn]

http://www.isuppli.com/Memory-and-S...byte-NAND-Resuscitate-Solid-State-Drives.aspx

After two years of inflated prices, NAND flash memory at the end of 2010 is set to return to the key $1 per gigabyte level considered a key threshold to drive adoption of Solid State Drives (SSD), but the cost reduction may come too late to help the struggling SSD market, according to the market research firm iSuppli Corp.

Pricing for 1Gbyte of 3-bit per cell (TLC) NAND flash memory will average $1.20 for the entire fourth quarter and then decline to $1.00 at the end of the year. This represents a precipitous drop from the first quarter of 2010, when pricing for TLC averaged $1.80 per gigabyte and 2-bit per cell (MLC) flash was at $2.05, iSuppli memory pricing and forecasts indicate. It also marks the first time NAND flash pricing fell below the $1 threshold since the fourth quarter of 2008, when MLC pricing averaged 90 cents per gigabyte.



NAND Flash Pricing

“When NAND pricing first fell below the $1 level at the end of 2008, many observers opined that its would sound the starting gun for solid state storage, allowing the technology to be cost competitive with Hard Disk Drives (HDDs) in PCs for the first time,” said Michael Yang, senior analyst for memory and storage at iSuppli. “However, during the following quarters, pricing rose because of strong demand and constrained production capacity, limiting the appeal of SSDs to low-volume servers in data centers and preventing widespread adoption in high-volume business and consumer PCs.”

Now with pricing back to the $1 level, the SSD market is ready to get back on track—or is it?

“With NAND pricing having returned to per-gigabyte pricing levels not seen in two years, there’s likely to be a lot of new buzz created for the solid state storage market at the end of 2010,” Yang said. “However, traditional HDDs gained a lot of additional ground during the past few years in terms of rising capacity and falling prices. In fact, HDDs have gained so much ground that SSDs now are in danger of never regaining their competitive footing.”

To compete successfully with HDDs, per-gigabyte pricing for NAND flash memory will have to decline to 40 cents by 2012, Yang opined. At such pricing, a 100Gbyte SSD could cost $50, when supporting electronics are added in. This would make solid state storage more appealing in consumer and corporate PCs.

A Little TLC Goes a Long Away

The second half of the year is almost certain to see shortages for MLC NAND. However, the complete opposite is the case with TLC as supply is sufficiently ahead of demand. Average capacity for TLC chips in SD (secure digital memory) cards and USB storage devices has stagnated during the past year. When combined with a slowing unit demand due to seasonality, ASPs for TLC chips are falling.

Reaching Consumers

Low TLC prices are reaching consumers, which will aid in the purchase of higher-density products. iSuppli predicts the third quarter will see an uptick in growth for TLC NAND flash memory—driven by numerous holidays and back-to-school promotions.

To this end, tightly managed inventory levels, along with the closing of some factories, are paving the way for new price competition in the TLC segment, indicating a very real possibility that the magical $1 per gigabyte level might be reached by the end of 2010.

That being said, iSuppli believes these lowered component prices, paired with seasonal demand growth in 2010, will lead the overall industry for NAND flash memory to reach a record $5 billion in revenue in the third quarter.

 

[uli2000]

At that price, I can see myself buying a small (~30-80gb) SSD for a boot drive and a few applications, but I don't see myself replacing hdds anytime soon.

 

[anishannayya]

I have two SSDs, but will replace all mechanical storage in my PCs when a 1TB SSD hits $250. I would keep the mechanical storage in my media server though, until SSD prices become as dirt-cheap as HDD prices

 

[pitz]

With the excellent iSCSI capabilities in Windows, and now, more BIOS' supporting iSCSI boot, it seems that the concept of having computers with hard drives, other than servers or NAS machines, is slowly dissappearing.

 

[VirtualLarry]

I don't get it. the blurb talks about SSD pricing, and then goes on about TLC NAND, meaning three-level-cell, which to my knowledge isn't used in SSDs yet, because of longevity concerns. (Which are very valid, IMHO. Ten write cycles per cell is not good.)

 

[wirednuts]

ssd's are like oled's. hell, even like lcd's. it took lcd monitors FOREEEEVVVERRRR to get affordable and mature on desktop applications. the same is happening with oled's. it will happen, its just happening realllly slow.

ssd's might still have another decade before they really start shoving hdd's to the side, but once it happens look out. the only moving parts in our pc's will be the fans, that is certain.

even better will be when ram technology will be able to retain its memory when the power is off. i truly believe ram and sdd's will be one in the same at some point, and motherboards will just start coming with 500gb or so of ssd/ram chips on them (like video cards do now, it will share part of the memory to serve as the ram, adjustable at will).

 

[anishannayya]

I don't think so. SSDs are much, much slower than RAM is.

If you were to take a computer with an SSD and replace its maybe 4 GB of RAM with 1 GB, then it would experience micro-freezes anytime you tried to do anything with it. Not to mention how quickly the flash cells on the SSD would die with so many writes.

RAM barely takes up electricity to be powered. On my laptop with 8 GB RAM and an 8-cell battery, I can leave the thing in sleep mode all night and not even loose 1 percent of my battery life.

 

[Fayd]

what a silly article.

harddrives are not direct competitors to SSD's. SSD development hasn't stalled. rotational harddrives will NEVER be faster than SSD's.

 

[Hacp]

what a silly article.

harddrives are not direct competitors to SSD's. SSD development hasn't stalled. rotational harddrives will NEVER be faster than SSD's.

Rotational hard drives don't need to be faster than SSDs. They just need to eliminate the laggy feeling.

 

[anishannayya]

Rotational hard drives don't need to be faster than SSDs. They just need to eliminate the laggy feeling.

Oh, you mean, like, by getting faster!

 

[Idontcare]

but the cost reduction may come too late to help the struggling SSD market, according to the market research firm iSuppli Corp.

Honestly now, when was the last time iSuppli was right about anything?

They're business model seems to be centered around intentionally selling bad forcasts and wrong market projections simply so they can come back in 3 months and sell you an "updated" version that is less wrong than before.

The ones that kill me is when they come back say 3 months after the fact and sell a "revised" marketshare projection for a year that has already passed (for example, come March of 2011 they will finally publish their official projections for 2010).

 

[Hacp]

Oh, you mean, like, by getting faster!

Rotational Hard Drives will always slowly improve their sequential speeds because of data density. The only area they need to improve on is random read/writes. There must be a good reason why hdds haven't switched to a log based system with a large read cache for small files. SSDs use that system and it seems to work fine.

 

[sxr7171]

Hogwash. SSD is the future of all storage. I have my wallet in hand for $1/GB (+20%) SSDs and I'm sure I'm not the only one. Rotational disks will be only for mass storage in 5 years if that. So they had a small setback in early 2010, it won't change things in the long term.

 

[anishannayya]

Rotational Hard Drives will always slowly improve their sequential speeds because of data density. The only area they need to improve on is random read/writes. There must be a good reason why hdds haven't switched to a log based system with a large read cache for small files. SSDs use that system and it seems to work fine.

Remember that thing called physics? Well, that's what will cause us to hit the limit on data density for rotational media.

In regards to random reads, physics is again the culprit. You can't randomly read anything when you have a mechanical device. The only way to combat randomness is with instantaneous access; Random = instantaneous. Having to move a drive head isn't the best way to perform an instantaneous action. Couple that with having to spin the platters a couple thousand times per minute so the drive head can hunt for the data, and things start to get complicated and slow.

HDDs are out. Like every technology, SSD prices will hit the floor and saturate the market. Bigger sizes will come out, and OSs will begin to optimize for them. Don't hold your breath for the next HDD revolution; I guarantee that you'll pass out.

 

[randomlinh]

With the excellent iSCSI capabilities in Windows, and now, more BIOS' supporting iSCSI boot, it seems that the concept of having computers with hard drives, other than servers or NAS machines, is slowly dissappearing.

heh, that is a LONG ways off. In the corporate environment, I agree with you. But at home? I don't know a single person that has a "server" for central storage location. Other than myself. And even I backed off that because I wasn't happen w/ the solutions at the time (I now don't need all that storage after I gave up on a video library. my photos don't eclipse 1TB yet, so it's cheaper to just have multiple external drives for backup).

What will more likely happen to the masses is everything will be in the cloud. Our home computers will become more and more like appliances that connect to itunes, google, or microsoft. It seems as people don't care about data security. They trust it'll always be online. I welcome this, but still want to keep my own data. Not sold on the cloud as my sole source, but I'm not in the majority.

But back to topic, if it gets down to $1/GB, I will certainly willing to spend the money on a 160GB drive. It has to be at least a mainstream drive tho, not budget performance. Maybe an 80GB boot/apps drive and an 80GB working drive. Not sure if it's worth splitting up. But bring it, I'll buy it.

 

[FishAk]

I don't think so. SSDs are much, much slower than RAM is.

You have to qualify the RAM type: ie SRAM, DRAM, SDRAM ect, because an SSD is already just another type of RAM. Granted- still not as fast as your systems main memory... yet.


On the other hand, over the next decade or so- when your SSD is as fast the system memory you have now- your system memory may be so fast it'll have tomorrows weather already loaded.

 

[adamantinepiggy]

That article is bullshit. 3-bit MLC NAND is not going to be put into any SSD's anytime soon. Why? because it's slow crap at it's current evolution. Now maybe in the future it might, but not anytime soon. At this point it's good enough for USB storage stick like memory (slow and unreliable), but that's about it, and even then, it's not as fast or reliable as what is already being used. Yes memory stick memory will be able to grow in size with 3+ bit NAND, but they will be definately a lot slower for the foreseeable future.

Extra bits means extra slow and unreliable at this time. Plus any additional bit levels requires more error correcting, slower writes, etc, etc. Like I said, maybe later on down the line they might get it workable for "SSD" level applications, as opposed to storage stick type memory, but that aint happening real soon.

 

[Hacp]

Remember that thing called physics? Well, that's what will cause us to hit the limit on data density for rotational media.

In regards to random reads, physics is again the culprit. You can't randomly read anything when you have a mechanical device. The only way to combat randomness is with instantaneous access; Random = instantaneous. Having to move a drive head isn't the best way to perform an instantaneous action. Couple that with having to spin the platters a couple thousand times per minute so the drive head can hunt for the data, and things start to get complicated and slow.

HDDs are out. Like every technology, SSD prices will hit the floor and saturate the market. Bigger sizes will come out, and OSs will begin to optimize for them. Don't hold your breath for the next HDD revolution; I guarantee that you'll pass out.

Random reads are usually small reads. As I said, a large nand cache will solve the random read problem.

 

[Fayd]

Random reads are usually small reads. As I said, a large nand cache will solve the random read problem.

there already exist such hybrid drives that use a large nand cache in combination with rotational media.

they are better than pure rotational media for general functions, but they don't offer anywhere near the performance of pure SSD's.

stop being dumb.

 

[Cerb]

You have to qualify the RAM type: ie SRAM, DRAM, SDRAM ect, because an SSD is already just another type of RAM. Granted- still not as fast as your systems main memory... yet.

When speaking in general terms of RAM, one is almost always talking about some kind of of DDR SDRAM. So far, though, SSDs still aren't up to good old PC66 (we're almost there, though!).

Random reads are usually small reads. As I said, a large nand cache will solve the random read problem.

The drive needs to move a physical arm to the location it must read. Period. Some drives are good at ordering small amount of outstanding requests so as to do this very well (Samsung F3, FI), but that only goes so far, as does the drive's data cache (part of why big drive caches are still typically only 64MB). I can see how a flash cache could help in random R/W benchmarks, like IOmeter. But, in normal use, I fail to see how it will help that much, unless the cache is made so large as to be cost-prohibitive (30+ GB for a 1TB drive, maybe?).

Though, it appears that actually, even IOmeter results are not great. It seems to primarily help with non-random loads, as expected. Not that the technology is bad or anything, and it's certainly a good way to handle netbook-class machines, where you'll only have one SATA drive. But, it can't overcome the need to physically move the head, in the corner cases (the ones where your PC feels slow), where SSDs shine like the sun.

As to the article in the OP, I was ready to get out my popcorn as soon as I read, "resuscitate." SSDs are being adopted by anyone who can afford them now, and as the prices go down, all of us who can't justify the cost are just waiting for it to hit magic price points.

When quality 160-300GB SSDs get into the $75-$150 range, it'll be on. 160 is enough OS, apps (including many modern games), and documents waiting to be moved to a big HDD. 300GB or so will be enough the overwhelming majority of users, for whom multiple drives would add more complexity than they can manage.

 

[Voo]

there already exist such hybrid drives that use a large nand cache in combination with rotational media.

they are better than pure rotational media for general functions, but they don't offer anywhere near the performance of pure SSD's.

stop being dumb.

No reason to insult anyone, especially considering that the "hybrid drives that use a large nand cache in combination with rotational media" use 4(!)gb of SLC flash... which imo is everything but large... and does still rather good depending on the benchmark.
But still if we get really large amounts of flash cache, I don't see why we couldn't get the same result from a intelligent OS/controller and two independent drives. HDDs all in all are limited by their rotational speed, but for mass storage that's fast enough and I just don't see SSDs anywhere near 0.05€/gb in the near future.

In the long term SSDs have all the advantages on their side, but that will take some time, till then I'm completely fine with a SSD for my OS/apps and HDDs for anything else - my 160gb G2 still cost more than my 6TB of conventional storage

 

[Idontcare]

HDDs all in all are limited by their rotational speed, but for mass storage that's fast enough and I just don't see SSDs anywhere near 0.05€/gb in the near future.

true but don't worry too much as you'll see 0.05€/gb flash in about 4-5 yrs at current technology pace...however the point stands which is that rotational media will always have a cost structure that is an order of magnitude cheaper than flash technology.

Making flash will simply never be as cheap as stamping a platter. But the day is rapidly coming where it will be cheap enough that 90% of the market won't care what the price delta is, just like LCD vs CRT, because the absolute price for SSD will be below the threshold of caring.

People still use tape for storage and CRT displays for specialized applications, spindle discs won't disappear but their TAM will be cannabalized by SSD and the user-base for spindles will shrink dramatically.

 

[bunnyfubbles]

I don't get it. the blurb talks about SSD pricing, and then goes on about TLC NAND, meaning three-level-cell, which to my knowledge isn't used in SSDs yet, because of longevity concerns. (Which are very valid, IMHO. Ten write cycles per cell is not good.)

Honestly that wouldn't matter much to me as I already use large drives almost strictly for permanent storage, so a large drive that can't stand up to a lot of writes won't matter if I'm only writing a bunch of files to it and then only reading from it from then on out.

Obviously I would still want a separate and more robust boot drive, but if they can create a cheap storage SSD that is massive in size and cheap because it wouldn't be good for OS/apps and strictly for storage, it would be perfect for replacing these massive HDDs in an HTPC/home server role. Even though these "green" HDDs are quieter/cooler/more power efficient, an SSD still trumps them in all areas except for cost per capacity.

 

[Hacp]

The drive needs to move a physical arm to the location it must read. Period. Some drives are good at ordering small amount of outstanding requests so as to do this very well (Samsung F3, FI), but that only goes so far, as does the drive's data cache (part of why big drive caches are still typically only 64MB). I can see how a flash cache could help in random R/W benchmarks, like IOmeter. But, in normal use, I fail to see how it will help that much, unless the cache is made so large as to be cost-prohibitive (30+ GB for a 1TB drive, maybe?).

Which is why a big nand cache helps drastically with random reads. If you look at rotational hdd performance, the read numbers are good from 32k and up, but if you go below that, the reads are slow. That is probably due to the drive thrashing around trying to find the data. If you have a cache that just caches all the data below 32k you go a long ways to improving the feel of hdds. That is of course if you also move towards log-structured hdds that can hold data in a small cache then write the info sequentially.

 

[Hacp]

there already exist such hybrid drives that use a large nand cache in combination with rotational media.

they are better than pure rotational media for general functions, but they don't offer anywhere near the performance of pure SSD's.

stop being dumb.

As I said before, you only need to remove the laggy feeling hdds have to satisfy most people. The seagate momentus is flawed in that it does nothing to improve random write speeds, and also it doesn't use the newer 500gb platters which improve sequential throughput.