SSD Write Endurance 25nm Vs 34nm

[Christopher]

Alright -- so I know that one or two of you have experience with the MTRON PRO 7000 -- or at least something like it.

Would it be feasible to test it in a reasonable amount of time? MTRON claims you could write 50GB a day for 140 years. That is roughly 2.5PB, and it is not very fast.

I won't have it here for a few more days, but I don't know what else to do with a 16GB MTRON other than endurance test it.

[bulanula]

Does this 5200 hours or 5000+ hours issue also affect the C300 or just the M4 ?

[Ao1]

Still waiting for my 830 to be delivered. Someone provided data on an 830 used in a production environment, which can be compared to the output from Christopher’s drive. (The 1st table is Christopher’s drive; the 2nd is the production drive)



Uploaded to ImageShack on account of XS not allowing you to upload images ImageShack.us

I’ve been hassling Samsumg for info on the 177 value. What they have stated in item 2 & 3 below does not make sense and I don’t think I will get any further clarification.

1) Attribute 177 can update while the unit is on
2) The raw value on this attribute is how many time it can prolong the life of a specific block
3) The indicator range from a wider range of numbers as 1-100 is not enough.

Data from the second table seems to show that 177 can update without a power off. Maybe that is due to the different workload or maybe it just a configuration quirk on Christophers set up. Let’s see what happenes with my drive.

I still can’t be sure that 177 is the MWI, but for the purpose of the endurance test I’m going to assume it is. When it gets to 1 I will stop the test, disconnect the power and leave it for 3 months. I’m going to run a few trace benchmarks and then I will let the standard endurance test run for a day or so. I will then ramp things up with the 4K random workload.

[Christopher]

Thanks for the chart Ao1.

For the record, the 830 is being tested on an H67. Initially, I was using the RST drivers... but after the "Great Samsung 830 Performance Apocalypse of 2011" I switched to MSAHCI. I had no idea that 177 was at 1 because I didn't power off the drive for over 10 days. 177 data after 10 days was up to ~130... then after a power cycle it was over 5400+. So I saw it go from 96/0/130ish to 1/0/5400ish+ after turning the system off then back on.

Also remember -- and I can't stress this enough -- that the drive is not working as it should, at least performance-wise. It's performing more like a 64GB C300 than 830, so there is no telling what other issues may be lurking. I'm thinking about getting another one.

Are you getting the 64GB 830 in the mail? (or is it a different capacity?)

[johnw]

I still can’t be sure that 177 is the MWI, but for the purpose of the endurance test I’m going to assume it is.

If you fit a line to the 177 RAW vs. 177 normalized values, the slope comes to about 30 raw per 1 normalized. And 100 * 30 = 3000. So it seems likely 177 is counting down, in normalized units, from 3000. Which is likely the expected number of cycles for the flash.

For comparison, with the Samsung 470, normalized 177 attribute hit 1 when the raw value was about 5017, so it seems like the 470 was using flash with an estimated 5000 cycle lifetime.

It is interesting that the WA comes out less than 2 for the "production environment". I wonder if the reason we measured 4 or 5 with the endurance test is because with continuous writing the Samsung never does much GC and so ends up having to do a lot more block erases. But maybe in the production environment, there are sufficient pauses for Samsung GC to kick in? Just a guess.

[Anvil]

Kingston SSDNow 40GB (X25-V)

607.98TB Host writes
Reallocated sectors : 05 19
Available Reserved Space : E8 99
POH 5496
MD5 OK

33.98MiB/s on avg (~47 hours)

--

Corsair Force 3 120GB

01 94/50 (Raw read error rate)
05 2 (Retired Block count)
B1 63 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 644147 (Raw writes) ->629TiB
F1 857347 (Host writes) ->837TiB

MD5 OK

106.74MiB/s on avg (~95 hours)

power on hours : 2469

[Zaxx]

Quick (and old) question about a previous test of the Crucial M225 drive with Samsung 51nm nand. The P/E listed is 5000/10000...isn't all 51nm created equal or what am I missing?

I'm interested cuz I have an old Ocz Apex 64Gb (ex-os) drive that I now use as a scratch drive and it uses 51nm Samsung as well. It uses 2 of the infamous stuttering (w/XP) JMicron 602 rev. b controllers (they doubled the cache to a whopping 8MB) internally raided. I wish I knew the millage but the only smart attribute it supports is a fixed 44C temp reading. I hope I get even close to the writes of the M225 but I doubt the JM controller manages nand as efficiently as the Indi does.

[Ao1]

It is interesting that the WA comes out less than 2 for the "production environment". I wonder if the reason we measured 4 or 5 with the endurance test is because with continuous writing the Samsung never does much GC and so ends up having to do a lot more block erases. But maybe in the production environment, there are sufficient pauses for Samsung GC to kick in? Just a guess.

That is what I would guess as well. “Normal” client use might be a completely different picture. It would be helpful if anyone reading this thread, that is using a 470 or an 830 in a client environment, could post their SMART data.

@Christopher, I’ll be using a Z68, which has Marvel for SATA III and an Intel chip for SATA II. I’m leaning towards using the Marvel controller. I’ve ordered the 64GB version, touch wood it will arrive tomorrow morning, otherwise I’ll have to wait until Monday.

[B.A.T]

Todays update. The power was cut for 10 min this morning and I didn't notice until after work. Everything is back up and running as normal.
Kingston V+100
323.4132 TiB
1753 hours
Avg speed 24.70MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
208,1707 TiB
20061 hours
Reallocated sectors : 00
MWI=144 to 143
MD5 =OK
50.56 MiB/s on avg


m4
185.8111 TiB
681 hours
Avg speed 78.16 MiB/s.
AD gone from 250 to 248.
P/E 3268.
MD5 OK.
Reallocated sectors : 00

[Christopher]

@Christopher, I’ll be using a Z68, which has Marvel for SATA III and an Intel chip for SATA II. I’m leaning towards using the Marvel controller. I’ve ordered the 64GB version, touch wood it will arrive tomorrow morning, otherwise I’ll have to wait until Monday.

Not that I think it matters much, but what Z68 board has no Intel SATA III ports on it? Some Z68 boards use Marvel for two extra 6gbps ports in addition to the two Intel 6gbps and for the purposes of testing it shouldn't matter really, but I am kinda curious as why Z68 SATA III ports would get axed in any 1155 board. Do you get two extra SATA II ports or something?

[Anvil]

There are 2 native SATA III ports on every Intel n67/Z68 motherboard, there is a Marvell controller on most Z68s as well but mostly it sucks for RAID usage, it can do pretty OK for single drives.
(there are a few exceptions to the Marvell controller performing badly, just can't remember which one it was 9182 or 9128)

[Christopher]

As far as I know, all 830s are using the same firmware, so I'd be curious as to see what happens with the new 830. I was contemplating buying another one myself.

[Christopher]

Samsung 830 64GB Update, Day 31
FW:CXM01B1Q

GiB written:
1917196.90

Avg MB/s
64.88

PE Cycles
12503, up from 12167 yesterday

Reallocated Sectors
40960
20 Blocks, up from 18

755 Hours






Here is a bonus pic, depicting the current state of performance of the 830:


Yes, it really is working in SATA III.

[Christopher]

This is what it was like before the performance fell off a cliff:


I think I'll try SEing it again... but I expect no miracles this time either. I could RMA it, but I'm not sure that the fundamentals related to this test are affected, other than the speed of the 830's demise.


On the other hand, I've been playing with another drive I am considering testing. I bought two similar drives, but both were acting a tad strange out of the box. Here is one of the drives:
Notice how similar it and the 830 are currently performing (see the 830 ASU shot in the post above, the last pic from today's update) in ASU...

[minpayne]

@Christopher

The Samsung 830 is quite interesting... I've been holding off from the 830 512GB since I've seen your posts regarding the drop of performance (which is yet recovered) and the write-amplification calculated by Ao1. What worries me most is the sudden and severe change of the MWI-like SMART attribute since the power cycle. Back to Intel 320 Series for now and I don't plan to touch new controllers for a laptop used for production.

Just curious, have you verified that the motherboard, the SATA controller and the cables are still working correctly, i.e. you could reproduce SATA-III speed with a "good" SSD?

[minpayne]

@Ao1

The log of the production drive in your chart could probably tell that the B1/177 serves as MWI and WL-count at the same time. If fitting the raw value against the value with a simple linear regression then it pretty matches the assumption that the expected P/E is around 3000 when the MWI is exhausted.

[Christopher]

@Christopher

The Samsung 830 is quite interesting... I've been holding off from the 830 512GB since I've seen your posts regarding the drop of performance (which is yet recovered) and the write-amplification calculated by Ao1. What worries me most is the sudden and severe change of the MWI-like SMART attribute since the power cycle. Back to Intel 320 Series for now and I don't plan to touch new controllers for a laptop used for production.

Just curious, have you verified that the motherboard, the SATA controller and the cables are still working correctly, i.e. you could reproduce SATA-III speed with a "good" SSD?

Look, it's pretty much impossible to get this to happen naturally. I made some changes to the amount of free space after the second day, as the drive had become quite slow. I bumped it back up where it was, and it was fine for another 8 days or so... until I reduced the amount of free space left again. I let it run longer to see how bad performance would get, and this time secure erasing it wouldn't even fix it, much less the more extreme methods I tried.

I can put the drive into another system, and it does the same thing in every system I've tried, and with different drivers as well.

If this was a widespread problem, Dell, Lenovo, and Samsung would have stumbled upon it already.

[minpayne]

Look, it's pretty much impossible to get this to happen naturally. I made some changes to the amount of free space after the second day, as the drive had become quite slow. I bumped it back up where it was, and it was fine for another 8 days or so... until I reduced the amount of free space left again. I let it run longer to see how bad performance would get, and this time secure erasing it wouldn't even fix it, much less the more extreme methods I tried.

I can put the drive into another system, and it does the same thing in every system I've tried, and with different drivers as well.

If this was a widespread problem, Dell, Lenovo, and Samsung would have stumbled upon it already.

OK, what we could do now is to wait for more samples of the 830 to be tested. I just don't understand why the WA had always been exactly 0.1 before that power cycle. I guess something is not quite right with the firmware (especially when under such stress).

[Christopher]

The WA was .1 because 177/Wear Leveling was not being reported correctly... it only increased by 130 the first 10 or so days, but then after I disconnected the drive, 177 reported correctly. It does not really increase on my drive while the drive is powered on... you have to manually power cycle it for Wear Leveling count to increase. For the past few days I've been power cycling the drive or system before taking the SMART screen shot.

[minpayne]

You mean the 177/WL-count was increasing, but not increasing correctly, while you were not power cycling it? This is what looks strange to me.

[minpayne]

OK, here's my speculation of what happened to your 830:

Assume that the P/E spec is 3000. Assume that the true WA is 4 under ASU stress. The expected endurance for host writes is: 64GB/(1.024^3)*3000/4 = 44703GB

In such case, could it be that the 830 degraded fast after you passed 44703GB of host writes?

[Anvil]

Unfortunately there is some crucial info missing from Christophers run as that exact attribute failed to update, lets hope that it not the norm.

[minpayne]

I have a question regarding the data retention test:

How to verify data retention? Do I set the disk to "Offline" in Windows Disk Management, then use WinHex to calculate an MD5 or SHA-1 of the entire physical disk, then power off the disk for, say, 10 days, then power it on and re-calculate the MD5/SHA-1? I worry that each time I connect and mount the disk, something will be written to the disk by the OS, hence resulting in different hash value of the entire disk.

Edit: just confirmed that Windows 7 would write something into the disk each time it's plugged, resulting in a different hash value of the entire disk. It looks like the data retention test would have to be done on a hash value of a large file instead?

[bulanula]

Eagerly awaiting some more C300 results. Hopefully one day it will beat that M4.

[Ao1]

At last it has arrived.





Here goes, I'm starting off with the normal endurance workload to see how things progress with the SMART data before ramping up to 4K random full span