Two Samsung 840 Pro's & 1 840 have failed, 4 at [H]ardOCP (Early review sample SSD's)

[Hellhammer]

Originally Posted by jwilliams4200

For the 40TB-written 840, have you tried any data retention tests? For example, copy a few very large files to the SSD, compute the MD5 checksums, then power off the SSD for a week. Then power it back on and check the MD5s...

No, I haven't done that (at least yet). The WLC (Wear Leveling Count) is only at 75 so I need to hammer it a bit more as currently the data retention should be much longer than a week (most manufacturers rate their MLC NAND so that the data retention time is 1 year after the P/E cycle count has been achieved). The data I have so far suggests that the WLC will hit 1 after ~1300 P/E cycles.

[taltamir]

Originally Posted by jwilliams4200

For the 40TB-written 840, have you tried any data retention tests? For example, copy a few very large files to the SSD, compute the MD5 checksums, then power off the SSD for a week. Then power it back on and check the MD5s...

1. Its supposed to retain power for 10 years. Testing a week is not useful.
2. MD5 checksum failure requires that sufficient cells would fail to the point where built in ECC cannot correct it (recall that all modern SSDs have ECC to compensate for their increased error rate). So even if some data was lost, it will not show in MD5 checksum unless a whole bunch of data was lost

[jwilliams4200]

Originally Posted by taltamir

1. Its supposed to retain power for 10 years. Testing a week is not useful.

Wrong. The data retention decreases as erase cycles increase. According to one common specification, the data retention should still be at least 1 year at the rated maximum erase cycles for the device. But if you surpass the rated erase cycles, the data retention time can become very short, potentially less than a week, or even less than a day.

In addition, the Samsung 840 uses TLC flash and I am not sure if Samsung adheres to that specification. Furthermore, Samsung has not specified the erase cycles lifetime of the flash in the 840, as far as I am aware.

[jwilliams4200]

Originally Posted by Hellhammer

The WLC (Wear Leveling Count) is only at 75 so I need to hammer it a bit more as currently the data retention should be much longer than a week (most manufacturers rate their MLC NAND so that the data retention time is 1 year after the P/E cycle count has been achieved). The data I have so far suggests that the WLC will hit 1 after ~1300 P/E cycles.

This is the 250GB model that you are testing, right?

40TB / 256 GiB = 145.5

So the erase cycles should be at least 145.5, but probably higher since average write amplification should be greater than 1.0

I think some of the early Samsung 830s had a firmware bug where the WLC did not get updated reliably. I wonder if something similar is happening with your 840.

[Hellhammer]

Originally Posted by jwilliams4200

Furthermore, Samsung has not specified the erase cycles lifetime of the flash in the 840, as far as I am aware.

1,000 P/E cycles is what I heard/saw while I was in Korea. While the info was straight from Samsung, it isn't exactly official (I saw the figure during a fab tour).

Originally Posted by jwilliams4200

This is the 250GB model that you are testing, right?

Yup.

40TB / 256 GiB = 145.5

So the erase cycles should be at least 145.5, but probably higher since average write amplification should be greater than 1.0

I think some of the early Samsung 830s had a firmware bug where the WLC did not get updated reliably. I wonder if something similar is happening with your 840.

I should note that the WLC was at 89 when I started endurance testing. I had run all our regular tests on the drive and used it as a secondary boot drive for a few weeks before I started testing, hence it wasn't 100.

So far the WLC has been going down linearly and the results make sense. The reported Total LBAs Written are way off though (I've tried SSD Magician, GSmartControl, Intel SSD Toolbox and CrystalDiskInfo - all report different numbers and they just don't make any sense). Fortunately I can still get the host writes from the average write speed reported by Iometer; WA should also be close to 1.0 as I'm using incompressible 128KB sequential data (QD1).

It's of course possible that the reporting is off and to be honest I wouldn't be surprised if it was (you will find out why in a few weeks when we are allowed to talk about this one thingy).

[jwilliams4200]

Originally Posted by Hellhammer

I should note that the WLC was at 89 when I started endurance testing. I had run all our regular tests on the drive and used it as a secondary boot drive for a few weeks before I started testing, hence it wasn't 100.

Ah, you mean the normalized value (of attribute 177). Right?

I thought you were giving the raw value of attribute 177, which (at least on the 840 Pro) appears to indicate the average number of erase cycles for the flash.

[Hellhammer]

Originally Posted by jwilliams4200

Ah, you mean the normalized value (of attribute 177). Right?

I thought you were giving the raw value of attribute 177, which (at least on the 840 Pro) appears to indicate the average number of erase cycles for the flash.

Yeah, I've been recording the normalized value. Didn't know that the raw value might indicate the number of erase cycles, I'll start recording it now to see what it gives us. It's at 250 now, which would mean 1,000 P/E cycles since the normalized value is 75.

[taltamir]

Originally Posted by jwilliams4200

Wrong. The data retention decreases as erase cycles increase. According to one common specification, the data retention should still be at least 1 year at the rated maximum erase cycles for the device. But if you surpass the rated erase cycles, the data retention time can become very short, potentially less than a week, or even less than a day.

1. Do you have any idea how long it takes to exceed the PE limit on a drive even when that is what you are intentionally attempting to do?
2. You started with a "Wrong." mr genious, but how does the "facts" you provide actually contradict my arguments?
3. I love how you conveniently ignore the parts of the argument you don't have a "refutation" (and I use that word generously) for.

[jwilliams4200]

Originally Posted by taltamir

1. Do you have any idea how long it takes to exceed the PE limit on a drive even when that is what you are intentionally attempting to do?

Yes, of course. I've done it.

[jwilliams4200]

Originally Posted by Hellhammer

Yeah, I've been recording the normalized value. Didn't know that the raw value might indicate the number of erase cycles, I'll start recording it now to see what it gives us. It's at 250 now, which would mean 1,000 P/E cycles since the normalized value is 75.

Good, it looks like what you heard in Korea agrees with what attribute 177 is indicating -- 1000 erase cycles for the new TLC flash in the 840. So, for the 250GB SSD, it should take around 250 to 275TB of sequential writes to hit the erase cycle limit.

[taltamir]

Originally Posted by jwilliams4200

Yes, of course. I've done it.

1. On what drive, specifically
2. How long did it take, specifically

[jwilliams4200]

Originally Posted by taltamir

2. How long did it take, specifically

You should not need to ask me...this is hardly rocket science. But for the sake of the mathematically and logically challenged:

(secs to exhaust erase cycles) = (flash capacity of SSD in MB) x (rated erase cycles) / (best sustained sequential write speed in MB/sec)

[taltamir]

Originally Posted by jwilliams4200

You should not need to ask me...this is hardly rocket science. But for the sake of the mathematically and logically challenged:

(secs to exhaust erase cycles) = (flash capacity of SSD in MB) x (rated erase cycles) / (best sustained sequential write speed in MB/sec)

1. You have been rude throughout, stop with the put downs. Personal attacks are not allowed.

2. I didn't ask you for the equation, I know how to do math. I asked you how long it took YOU to do it.

You claim to have exceeded the PE endurance limit on a drive, not to have done the math. I want to know the exact make, model, and size of the drive you have exceeded the limit on, and I want to know how long it took in actuality (not in theory).

Also, the equation you gave? It cannot be solved without plugging in information which you are REFUSING to give. Such as the make and model of the drive in question.

[jwilliams4200]

The SSDs I have worn out are not relevant to this thread. I'm not sure why you are so obsessed with me and my equipment.

Is it really so difficult for you to look up the sequential write speed of an SSD you are interested in, and then do some simple arithmetic using the formula I already provided to you? I know you can count at least to 3. I thought that anyone posting in this thread should be capable of basic arithmetic.

[katkah]

Originally Posted by jwilliams4200

Why do you keep posting nonsense? Are you intentionally trying to make yourself look like you have no idea what you are talking about?

The first failure was reported in the 840 Pro review when the review came out in late September.

I read the review just before I purchased on 11/06/12 and there was no fail written in that report! I know how to install benching software and hit start too! Read too! Show me a google cache of the report that says failure prior to the date of this posting! YOU CAN'T because it doesn't exist.

[Hellhammer]

Originally Posted by katkah

I read the review just before I purchased on 11/06/12 and there was no fail written in that report! I know how to install benching software and hit start too! Read too! Show me a google cache of the report that says failure prior to the date of this posting! YOU CAN'T because it doesn't exist.

There has been a mention of the failed unit since the review was published. Now the review has been updated with a mention of a second failed unit.

[jwilliams4200]

Not only did your reading comprehension fail on that article, you do not understand how google cache works. It only caches one copy of a page, the most recent one that it has successfully indexed. Since the review was updated when the second drive failed (and google indexed it), that is the one in google's cache.

The Internet wayback machine does not have any version of the article stored.

However, all you have to do is look at the article's comments, several of which mention the failure. For example:

http://www.anandtech.com/show/6328/s...eview?all=true

Review sample death? by serpretetsky on Wednesday, September 26, 2012
"Assuming the premature death of our review sample was a fluke and not indicative of a bigger issue,"
Care to elaborate on this? It simply failed after you guys ran all the tests on it?
Is there any way you guys might get another sample and run it through the same (or harder) tests?

Thanks for the review, you guys are awesome.

[Engineer]

Originally Posted by katkah

I read the review just before I purchased on 11/06/12 and there was no fail written in that report! I know how to install benching software and hit start too! Read too! Show me a google cache of the report that says failure prior to the date of this posting! YOU CAN'T because it doesn't exist.

If you're talking about Anand's review of the 840 Pro, he did indeed have that the first drive failed and a replacement was either on the way (or already received...don't remember which). It was in the review when it was published because I remember reading it and thinking it might be good to hold off on this before making a decision.

[zero koopa]

well I am glad I just saw this article after I ordered from new egg

[iaco]

Originally Posted by Hellhammer

Anand had both of the drives so I'm not sure if he did that kind of testing. He told me he's suspecting that it's the power delivery circuit but without a confirmation from Samsung, it's just a guess.

DIPM being disabled could be a factor though it wouldn't explain why the non-pro is fine.

[jwilliams4200]

Originally Posted by iaco

DIPM being disabled could be a factor though it wouldn't explain why the non-pro is fine.

I don't follow. DIPM being disabled where, exactly?

And nice de-lurk!

[iaco]

Anand's review of the 840 Pro showed much higher power consumption than Samsung's numbers. On page 3 of the comments section of the review, someone noted that DIPM not being enabled would explain the discrepancy.

If the drive is using more power than the power circuity was designed for then failure is possible.

[Hellhammer]

Originally Posted by iaco

Anand's review of the 840 Pro showed much higher power consumption than Samsung's numbers. On page 3 of the comments section of the review, someone noted that DIPM not being enabled would explain the discrepancy.

If the drive is using more power than the power circuity was designed for then failure is possible.

DIPM is never enabled on desktops by default, it's a feature limited to laptops. In the actual datasheet (which we got after the 840 Pro review went live) Samsung reports DIPM and non-DIPM numbers, which are on-par with our testing. It's very unlikely that the failure is related to this as that would be very poor engineering.

[Mr. Pedantic]

Originally Posted by iaco

DIPM being disabled could be a factor though it wouldn't explain why the non-pro is fine.

Why does a drive failure need to be 'explained'?

[serpretetsky]

Originally Posted by Mr. Pedantic

Why does a drive failure need to be 'explained'?

I'm not sure i'm understanding your question. Are you asking why it's important to find out why a design is failing in the real world?