backtop

Print E-mail del.icio.us 23 comment(s) - last by michael67.. on Oct 5 at 2:42 PM
Poor quality controllers to blame, says Samsung

DailyTech reported several days ago that SSD manufacturers have been having problems with Samsung's latest generation of 32nm NAND flash. The new chips have slow write speeds, thus making them unsuitable for use in SSDs since that is one of the major advantages that the new drives have over conventional hard drives. 

We have received a reply from Samsung confirming the issue, stating that "for quality SSDs, every NAND process geometry upgrade requires a matching upgraded controller.  Should (Samsung's) 30nm-class NAND be used with a conventional controller of insufficient quality, performance slowdowns are indeed possible".

Flash memory stores information in arrays of memory cells made from floating-gate transistors. As these transistors scale to smaller process geometries, it becomes harder for electrons to flow. This sometimes causes errors in writing to memory cells which must be corrected through ECC (Error Correcting Code). ECC is typically handled on the flash controller, which may be overloaded by excessive write errors if it is not sufficiently powerful enough. This is the most likely scenario for what is happening.

Companies we spoke with confirmed similar problems with 32nm flash from Toshiba that had been overcome. Intel is using 34nm flash from IMFT that was delayed from mass production for six months, possibly due to a similar problem as well.

Most of the SSD manufacturers we spoke with had paired Samsung's flash with Indilinx's Barefoot flash controller. There are several iterations of the Barefoot controller out there for different SSDs, and no doubt Indilinx is working on the problem. However, it might take a while, and sharp price drops on SSDs are unlikely for several months.

Meanwhile, Samsung is currently in the process of completing a new flash controller revision for their own line of SSDs, and have not released any SSDs of their own using the new flash memory. Samsung states: "We spend many months developing and then fine-tuning the controller and firmware technology for our SSDs, working very closely with most of the major PC OEMs".

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
Errors?
By Scabies on 10/2/2009 12:55:14 PM , Rating: 2
quote:
ECC is typically handled on the flash controller, which may be overloaded by excessive write errors if it is not sufficiently powerful enough. This is the most likely scenario for what is happening.

...
So its not that the new 32nm NAND is "slow," per se, but its just so faulty that it overwhelms the controller. Why am I hesitant to adopt?
Also, is the Samsung guy talking about 30nm in the quote mid-article a typo? Or does that translate to "in the thirties"?
quote:
30nm-class NAND be used with a conventional controller of insufficient quality, performance slowdowns are indeed possible




RE: Errors?
By TomZ on 10/2/2009 1:11:42 PM , Rating: 2
Most things don't work right or perform well during development. That part shouldn't scare you off. As long as there is good management, they won't release the product until it is performing well.


RE: Errors?
By The0ne on 10/2/2009 3:24:53 PM , Rating: 2
I know people are excited about SSD but I always advise consumers to wait for it to mature. Competition is good and prices have fallen in addition to increase performance. But quality will always suffer from the fast past. I'm expecting quite a bit from SSD after summer of next year :)


RE: Errors?
By TomZ on 10/2/2009 5:15:11 PM , Rating: 2
But by waiting, you are missing all the fun of being an early adopter. Sure, there can be some bumps along the way, but for many people it is an overall positive experience.


RE: Errors?
By dagamer34 on 10/3/2009 12:07:00 PM , Rating: 2
Fun stuff like watching your bank account get smaller from huge chunks bitten out of it? I'll pass.


RE: Errors?
By Master Kenobi (blog) on 10/3/2009 5:54:48 PM , Rating: 2
I will early adopt things like Intel processors, but never something like a SSD.


RE: Errors?
By dragunover on 10/4/2009 10:04:29 AM , Rating: 2
That makes no sense then.
If you get a 300-600 dollar investment that's a ~5-35% improvement why not get a 300-600 dollar investment that's 100-1000% better? Makes no sense to me...


RE: Errors?
By MatthiasF on 10/4/2009 1:22:57 PM , Rating: 1
SSDs don't give 100-1000% more performance.

Modern standard hard drives with higher densities, like inside the 1 TB and above models, are not that far behind SSDs for speed. Compare these benchmarks.

http://www.xbitlabs.com/articles/storage/display/c...
http://www.xbitlabs.com/articles/storage/display/1...

The only thing SSD has over normal hard drives are access times, but how helpful is that for most computer users?

The question should be asked is why someone is willing to pay $3 a Gigabyte for 30-50% faster transfer speeds and smaller hard drives, when a normal hard drive at $0.16 a GB performs just as well.

What we really need are HYBRID drives. Standard hard drives with NAND caches for areas we access often. This would improve speeds and be a moderate solution that has the best of both worlds.


RE: Errors?
By Seer on 10/4/2009 5:44:48 PM , Rating: 2
quote:
The only thing SSD has over normal hard drives are access times, but how helpful is that for most computer users?

:O
You really know nothing about typical home-computer usage patterns, do you? Random read/write, which is HIGHLY dependent on access times, is the most important measure of an HDD's performance.


RE: Errors?
By MatthiasF on 10/4/2009 6:29:16 PM , Rating: 1
Most important? Hardly. Access time is a negligible overhead margin for most hard drive operations. When the average file on a computer is 200 KBytes and bigger, a 5 ms access time is barely noticeable. The larger the average file, the less important the medium's access time.

Transfer rate is the most noticeable improvement from a storage medium.

Low access times really only benefit high IOPs situations that I doubt exist on a "typical home-computer".


RE: Errors?
By geddarkstorm on 10/5/2009 2:37:16 PM , Rating: 2
Really? You seriously believe that? Ok. Well, take a look at this then http://techreport.com/articles.x/17183


RE: Errors?
By michael67 on 10/5/2009 2:42:32 PM , Rating: 2
Until you install a SSD in your own system ore worked on one that has one you properly would never know how fat these little baby's are.

I have 3 vertex SSDs in my systems one 60GB as my system disk in my main system, One 30GB in my HTPC, and one 30GB i use between systems on a eSATA/USB ICYBOX that fits in a a 3.5" slot for games that i play on both PCs, so i can easily swap the disk between PCs.
http://www.raidsonic.de/en/pages/products/external...

Roughly boot and load times have halved, also auto-save's are less irritating.

Yes $500 is a lot of money on hard drives, but Rapors ain't cheap aider and not even half as fast as SSD (and yes i had a Raptor), and this has bin one of the most useful upgrades i ever done in price performance upgrades.

And o yeah there silent to!


RE: Errors?
By ElderTech on 10/4/2009 11:02:59 PM , Rating: 2
Unless you've actually used the latest generation of SSD, particularly those from Intel and OCZ, you can't imagine the improvement in overall performance over conventional hard drives. Due to environmental issues, I have one of the original Intel X25-M G1 versions that has been running flawlessly as an OS drive. Coupled with an i7 processor and other compatible components, it's simply amazing in speed and reliability. And it really makes a huge difference in the noise level of the system! For a comparison, I had originally purchased two Velociraptor 300gb drives to use for the OS, but got rid of them when the Intel G1 arrived and blew them away, both in performance increase and noise reduction.

Additionally, I also have recently added the Intel X25-M G2 drives to laptops with similar success. And at a price of $2.80/gb, it's well worth the money. If you really want to evaluate the current state of SSDs, by far the most technically competent and thorough research and reviews are at anandtech.com. Following is a link to his most recent review, but also be sure to check his archives for other relevant articles including extensive ones like his "anthology" on detailed SSD technology. Here's the link:

http://www.anandtech.com/storage/showdoc.aspx?i=36...


RE: Errors?
By afkrotch on 10/3/2009 4:52:23 AM , Rating: 2
Sounds like it'll just get worse and worse with each shrink, which a requirement of having to make more complex ECC to attempt to solve the issues.


RE: Errors?
By mindless1 on 10/3/2009 3:16:34 PM , Rating: 2
Uhh, no. ECC is not complex nor does it need be, they are simply shifting blame instead of accepting that if it writes slower, it writes slower.

What would help more is a larger DRAM cache, but when they only charge hundreds of dollars for an SSD, it would be too much to ask for $10 worth of DRAM. Cheap *bassturds*.


RE: Errors?
By fic2 on 10/2/2009 4:28:26 PM , Rating: 2
quote:
Also, is the Samsung guy talking about 30nm in the quote mid-article a typo? Or does that translate to "in the thirties"?
quote:
30nm-class NAND be used with a conventional controller of insufficient quality, performance slowdowns are indeed possible


I assume that 30nm- class refers to 30-39nm NAND chips.


RE: Errors?
By MatthiasF on 10/2/2009 5:11:06 PM , Rating: 4
One of two things could be happening.

First of all, the new NAND could require more than just EDAC for corrections, maybe also needing error scrubbing or perhaps even background scrubbing. If the memory controller itself isn't handling either, some of the tasks could be tossed up to the hard drive controller or even the CPU, slowing things down greatly.

Secondly, the new NAND being made by IM Flash Technologies (Intel/Micron) uses 3-bit per cell and not MLC, which needs a slightly different controller design.

Here's a helpful video explaining some of it.

http://www.micron.com/media/2009mediakit/3bitmlc_m...


IMFT Delay
By SAnderson on 10/2/2009 11:31:22 AM , Rating: 3
Post sources, original sources. Its not the first time I've seen DT mention IMFT's delays which never happened. Don't say look at the link in the article because that was not an original source of information. The last article didn't even mention how long the 'delay' was and now it all of a sudden its 6 months. DT is making sh1t up.




RE: IMFT Delay
By Jansen (blog) on 10/2/2009 12:00:24 PM , Rating: 2
Thursday, May 29, 2008

http://www.imftech.com/news/release_29may08.html

Based on the 34nm architecture, Intel and Micron also plan to introduce lower density multi-level cell products including single-level cell products, by the end of this calendar year.

There's more, including phone conversations I had, but I trust this is enough.


RE: IMFT Delay
By TomZ on 10/2/2009 12:28:39 PM , Rating: 2
How does that press release prove there was a delay, or that it had anything to do with controller issues as you implied in the article?


RE: IMFT Delay
By MatthiasF on 10/2/2009 1:41:59 PM , Rating: 5
He was an actual journalist and asked questions to Samsung and it's partners, finding out something that no one else has reported about openly?

It is possible, you know. Every source article on the Internet starts somewhere.


RE: IMFT Delay
By surt on 10/3/2009 11:46:10 AM , Rating: 2
quote:
Every source article on the Internet starts somewhere.


Yes, usually in someone's imagination.


RE: IMFT Delay
By SAnderson on 10/2/2009 3:38:40 PM , Rating: 2
That was a press release at which point customer samples were being sent out. At this point it's not in full production and not qualified to be sold yet, especially to a top tier customer that Intel probably is. Months go by before it does into full production, its the way the semiconductor industry works. Full production dates of the 34nm chips in fact do line up with the time frame mentioned in the IMFT press release. I repeat, there was no manufacturing delay.


"There's no chance that the iPhone is going to get any significant market share. No chance." -- Microsoft CEO Steve Ballmer

DailyTech Poll
Which web browser do you use on your primary personal machine? 






44 Comments









botimage
Copyright 2009 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki