Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
SSD help?
- Thread starter drummer1
- Start date
- Sort by reaction score
Don't get a Samsung 840. Just no.
It uses TLC NAND, which is lower reliability than MLC.
Samsung 840 Pro is okay though as it does not use TLC NAND. But it's too young of a controller for my tastes.
Personally, I'd go with the Vertex 3. The chipset has been out long enough that people are aware of pretty much any issue you'd have with it. But then again, I'm more cautious about data reliability than most people.
You are misinformed. The Samsung 840 does use TLC NAND and TLC NAND has fewer write cycles than MLC NAND (1,000 vs. 3,000). It has absolutely nothing to do with reliability.
See the chart below from this article.
So if you want to spend more money so your SSD lasts 70 years instead of 23 years (at 10GB writes per day), have at it. This is just a guess, but I doubt the OP plans to keep his computer 23 years.
Here is another Samsung 840 recommendation from forum user Hellhammer. He writes SSD reviews for Anandtech and knows more than any of us could ever hope to know on the subject. He is the author of the article I linked in my post above. If you read over a few of his reviews, you can see he really knows his stuff.
Honestly, with the newer SSDs, there are no bad SSDs IMO. Just get what works for you for the least amount of money.
I use an M4 in my MacBook Pro and I have no criticisms at all. It is flawless in both performance and price. Also, at least one part of me likes the fact it doesn't have Samsung written all over it
Of course TLC vs MLC vs SLC does have to do with reliability as the number of expected PE cycles is because of cell wear/tear. But additionally, TLC significantly increases the sensitivity to losing the stored data on reads too.
I've read the article before, and usually I take the Anandtech team's reviews/articles with high regard. But, I'll have to politely object to buying the 840 because I feel the review does not tell me much about reliability at all. Basically, the reviewer only verified that the WLC indicator is relative to 1000 cycles. It does not say anything about actual reliability aside than writing 92TB worked. I don't blame them, as there simply isn't that much raw data about TLC reliability over time, and they acknowledge this as they mention that vendors simply don't want to talk about it.
You're welcome to call me overly paranoid, but there's no way you could convince me to get a TLC drive over a MLC drive when the price is so close.
Besides, as others have suggested, there's plenty of MLC drives.
I'd agree with getting the m4 too. If they get it for $320, it's not even more expensive than the 840.
We will just agree to disagree about the definition of lifespan vs. reliability then.
Do you have a source for this?
OP scored a great deal. The cheapest I see the M4 online is $378.
IMO some people obsess a bit about SSD choices. There aren't really any bad SSD choices any more.
Last edited:
Didn't think a source was necessary as the phenomenon is well known in the industry and easy to explain to the general public. I haven't seen good stats of read cycles for TLC, but it's a given that it's worse than in MLC. But sure, I can google some for you too.
It's called read disturb.
Micron's presentation explains how it happens.
I also added in a link to NASA's mostly for a small piece of text on page 22 which says "Read disturb error rate much worse in parts that have consumed erase/program/read cycles." The PDF doesn't explain why, but it's because when a cell wears out, it's because the insulation around the floating gate wore out. And as the insulation thins, it becomes easier to influence a cell's value because there's less insulation.
Given that TLC is 8 voltage levels instead of 4 (MLC), you have even less margin against disturb effects.
Given that TLC wears out faster, you'll have a wider variation in disturb effect on neighboring cells to compensate for.
Given the latter compounds the former, more advanced and complex error correcting techniques are necessary to prevent disaster. If you've heard about the use of new DSP-style techniques in recent SSD controllers, TLC is why.
http://it.micron.com/~/media/Docume...em_summit_jcooke_inconvenient_truths_nand.pdf
http://nepp.nasa.gov/workshops/etw2012/talks/Tuesday/T04_Heidecker_Flash_Qualification.pdf
http://www.theregister.co.uk/2012/06/21/tlc_threshold/
Last edited:
No need for the condescension really.
Everything you posted explains that as a NAND cell reaches the end of its life if is more susceptible to errors. This will happen with any NAND type, MLC or TLC. So back to my original point, if you want to spend more money on a MLC NAND SSD so this does not begin to happen for 70 years instead of 23, go for it.
Post up any data you have on people losing data due to TLC NAND.
I believe the OP has the information needed to make an informed decision either way.
----------
Like I mentioned earlier, there really aren't any bad SSDs. Get what meets your needs for size and at a good price from a vendor you know will be there for you if you have trouble and need an exchange.
M4 for $315 is a good price I would say.
Go for it if you can get it for that price, otherwise they seem to cost the same as the newer and faster 480GB Crucial M500 lately.
I didn't mean to be condescending. Honestly. In the circles I normally discuss this in, this information is considered obvious.
But with regards to attitude on a forum post, as childish as it sounds, I feel you started it. I had written up a larger post which points out why I thought so, but I erased it because it's simply not worth spending time discussing our misunderstandings in the public. Sorry.
To answer your query: "Post up any data you have on people losing data due to TLC NAND."
I don't have any direct data that I can provide you on rates of user data loss from TLC. There's simply not enough consumers out there with TLC to get you a good sampling of long term real-world TLC SSD usage and failure rates. Any information I've seen regarding test data was confidential to the sources I discussed it with.
The most I can do with public information at the moment is provide you with the physics and operation of NAND, the differences between MLC and TLC, and show you how some of the differences makes it harder to reliably write/read data to a cell, and then how each of these differences compound with each other to approximate an exponentially worse uncorrectable bit error rate. Much of this has actually been summarized on Anandtech's SSD articles in the past, but those three links I posted have almost all of it with just a little more depth.
Yep. I have the 512 and 256 and both have been solid performers. At the time it was released, it was one of the top rated in many reviews. In a lot of the rankings it was only narrowly edged out by the Samsung 830, which I believe is the model Apple uses.