MLC vs SLC

"In fact, when compared to similar hard-disk drives, MLC NAND-based SSDs perform extremely well."

No kidding. ;-)

But SLC is dead; it won't be used anymore. The newer generation X25-E coming this christmas will use MLC memory as well; so this marks the end of SLC. By reserving more cells for internal usage (up to 50%) and create bigger SSDs, we improve the overall write iterations as well.

Still, i would have liked the limited writes to go away one day. I do a lot of benchmarking, but with my new 5 x Intel X25-V 40GB RAID0 array i won't be running random write benchmarks all day. Still, i think SSDs that reach their maximum write cycles is pretty rare; especially if it is a modern SSD with good wear leveling like Intel.
 
I posted this because it explained in a relatively simple way the difference between SLC and MLC. I know I didn't really understand the differences between the two types of SSD's, and that probably meant many other people didn't either. Yes, SLC is dead/dying, that was pretty much the point of the article. When you get the 5x X25e's LMK, I'll take those worn out, old school tech x25-v's out of your recycling bin! :D
 
Thanks for the link. Learned something new today about MLC

I had no idea that its now possible to have a semiconductor component with 4 different state or more :eek:.
 
I would hardly call SLC dead. It's still the only Flash tech which has the reliability and endurance required for commercial applications, such as in database caching. An MLC drive wouldn't last a month in such a situation.
 
Dead as in: not used anymore by major vendors. The X25-E was traditionally 'the' SLC product. But since the next-generation X25-E will use MLC memory instead; there won't be any high-performance SLC disks. Only slow controller-less SSDs like SuperTalent would use SLC still. Due to there being less flash cells, the write endurance isn't really higher than MLC disks anyway.

About realibility; how are MLC not reliable? Even if your product life halved; that wouldn't make it less reliable during its service life.
 
Hornet said:
Thanks for the link. Learned something new today about MLC

I had no idea that its now possible to have a semiconductor component with 4 different state or more :eek:.
Click to expand...

Well, it's 4 states per cell. I imagine the cells are made up of more than one transistor, obviously. It does say each cell holds 2 bits of information.
 
sub.mesa said:
Dead as in: not used anymore by major vendors. The X25-E was traditionally 'the' SLC product. But since the next-generation X25-E will use MLC memory instead; there won't be any high-performance SLC disks. Only slow controller-less SSDs like SuperTalent would use SLC still. Due to there being less flash cells, the write endurance isn't really higher than MLC disks anyway.

About realibility; how are MLC not reliable? Even if your product life halved; that wouldn't make it less reliable during its service life.
Click to expand...

MLC uses multiple values per cell, meaning that the possibility of a read error is more significant, which will only worsen as the cell degrades.

Further, 65 nm SLC has 100k writes/cell, whereas 34 nm MLC has about 3k. That's a lot of SLC cells and a lot of hacks to keep up the pretense of durability.

I can't see current MLC Flash being used for mission-critical systems.
 
Write iterations per cell are not all that important. You can have a million write cycles per cell; if you only have 10 cells that would mean 10 million write cycles. If you have 4000 write cycle cells but got a large bunch of them, the total number of write iterations would be alot higher; thus being able to handle more writes than the SLC cousin.

It's also the reason why using SLC is of no use anymore. You can get better write endurance by having 4 times the capacity and thus more total write iterations than a small SLC drive.

About your read error argument; that basically means SSDs should have read errors all over, as SLC SSDs are like 0.01% of the market and MLC like 99.99% - only thing i can read is failed OCZ drives due to firmware bugs. Didn't actually saw/meet/spoken to somebody who has an SSD with worn out MLC.
 
Elledan said:
Further, 65 nm SLC has 100k writes/cell, whereas 34 nm MLC has about 3k. That's a lot of SLC cells and a lot of hacks to keep up the pretense of durability.
Click to expand...

Correct regarding SLC, incorrect about MLC, though. The reality is about three times that value.

http://www.edn.com/article-partner/CA6319917.html

Also note that the article says, at 10,000 write/erase cycles, you could write and erase the entire drive once per day and it would still last 27 years. I don't think longevity will be a huge issue when HDDs hardly even last half that long.
 
That article talks about 90nm MLC NAND memory; 90nm!! Right now we are at 34nm for over a year and at christmas we're going 25nm.

The first generation Intel SSDs used 55nm SLC/MLC NAND. Currently they use 34nm so that's what we should be looking at.

Micron’s new MLC Enterprise NAND device achieves 30,000 write cycles – a 6x increase in endurance when compared to standard MLC NAND. And for enterprise applications that are more performance driven, Micron today also introduced a 34nm SLC Enterprise NAND device that achieves 300,000 write cycles – a 3x increase in endurance when compared to standard SLC NAND.
Source: http://www.desinformado.com/2009/10...vel-cell-nand-for-enterprise-storage-systems/

Still about factor 10 difference between MLC and SLC. However, as SSDs get bigger the total number of write iterations rise. And really; that's all you need. You better have a 1000GB SSD with 2.000 write cycles per cell; than a 100GB SSD with 10.000 write cycles per cell. One point against this way of thinking is that a larger drive might see more writes because it is used more intensively; since it holds more data people might use it also for other stuff like downloading, movies and other stuff that you normally put on HDDs.

I would be lying if i felt completely fine with the limited number of write cycles. One purpose i was thinking of (using multiple SSDs as ZIL/journal/log) is probably not a good idea; the performance increase is limited anyway and HDDs can do sequential write just fine.
 
You must log in or register to reply here.
Back
Top