SSD Write Endurance 25nm Vs 34nm

[johnw]

It seems you've got one more thing to play with the Samsung 830 There seems to be a new firmware released for it. Someone just mentioned that he updated his firmware version from CXM01B1Q to CXM02B1Q.

Hey, good find! Here is the link:

http://www.samsung.com/us/support/do.../MZ-7PC064D/AM

[Ao1]

Interesting. The timed workload MWI function are related to 226 & 227. It looks like the X25-E has these attributes, although they are not labelled as such. If they are timed workload attributes it would be easy to calculate how long it would take to deplete the P/E cycles.

IF Intel bring out a SF based drive presumably it will also have the timed workload function.

[minpayne]

Well, if the new firmware update doesn't cure the speed degradation of the Samsung 830, then I guess I tend to speculate that Samsung might have employed the so-called "throttling" strategy as rumored to be used by SandForce

[Ao1]

Samsung f/w update - it seems that atribute 177 can now update without a power cycle.

[minpayne]

Samsung f/w update - it seems that atribute 177 can now update without a power cycle.

Sounds great! Now could you please log it safely, and we might be able to find if its raw value is the average P/E count or the max P/E count or something else!

[Ao1]

This does not change what I have previously been able to report, although I can now report each incremental change in the dec value of 177.
This is the 4K random workload. I’ll switch to the client workload later when I have some time.

[minpayne]

Given that Micron claims to be able to secure erase an entire C400 (M4) 512 GiB within several seconds (officially within 60 seconds), I think it might actually be true that the nand writing speed can be a lot greater than the host writing speed, hence making the huge WA possible (e.g. 23x WA with 1QD 4k full span). Note that Samsung 830 has a "triple-core" controller.

[Ao1]

^

Correct, from a random Intel NAND spec sheet:

• Random read: 50 μs
• Sequential read: 20 ns
• Write performance
• Page program: 900 μs (TYP)
• Block erase: 2 ms (TYP)

So, at NAND level a block erase = 2ms

A block (in this example using Intel NAND) = 1MB

60 GiB = 61,440 x 2ms = 122,880 ms or 123 seconds

Host writes to fill a 60GiB drive @ 160MB/s would take 384 seconds

Let’s look at what the SMART data reports

In 10 minutes host writes = 2.54 GiB

If you multiple 2.5 GiB by the WA factor the writes at NAND level = 60.04 GiB

It appears entirely plausible that the combination of host writes and block erase times could be achieved within the time frame and I remain convinced that 177 is the MWI and the decimal value = 1 complete P/E cycle.

[Christopher]

First, there is no M2P, but there is a 64GB M2S (the earlier model, similar to the Corsair Performance 3s). It's slower, but basically the same thing.

Second, the X25-E is awesome, and I don't think you'll find too many people who disagree with that assessment (but it is not testing material -- unless you can wait 80 months for the results.)

Third, I have downloaded the new Samsung FW. It wasn't up when I checked yesterday, but I have the .ISO file so I will update the drive after todays' update, barring any objections from the peanut gallery. Wish me luck.

Lastly, I too was convinced that 177 was MWI as well, but trying to convince the SSDLife people of that didn't get anywhere. So my edition just has host writes and not MWI.

[Christopher]

Samsung 830 64GB Update, Day 35
FW:CXM02B1Q
Upgraded after update from FW:CXM01B1Q

GiB written:
222560.85

Avg MB/s
69.34

PE Cycles
14031, up from 13669 yesterday

Reallocated Sectors
40960
20 Blocks, up from 18

869 Hours




We will see what other effects the new FW has besides 177. I may SE the drive again to see if performance will go back to normal.

EDIT
Avg Speed is up quite a bit, even in the short time I have had CMX02B1Q up and running. It's increased 7MBs over the past hour.

UPDATE
I secure erased the drive, and left it idle for a few minutes. The good news is, speed seems to be back almost to where it should be. The bad news -- 177 doesn't seem to update. I still need to power cycle the drive to get 177's value.

[Anvil]

Kingston SSDNow 40GB (X25-V)

621.78TB Host writes
Reallocated sectors : 05 21
Available Reserved Space : E8 99
POH 5617
MD5 OK

33.81MiB/s on avg (~25 hours)

--

Corsair Force 3 120GB

01 77/50 (Raw read error rate)
05 2 (Retired Block count)
B1 67 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 678372 (Raw writes) ->662TiB
F1 902843 (Host writes) ->882TiB

MD5 OK

106.80MiB/s on avg (~25 hours)

power on hours : 2590

--

Looks like the Samsung FW update was a success, will do mine asap.

[Ao1]

Looks like mine is now worn in (Will confirm WA for the normal endurance load in a bit).

[johnw]

Looks like mine is now worn in (Will confirm WA for the normal endurance load in a bit).

Wow, 141.84 MB/s, I think that is a new ASU record for a 64GB SSD.

[Anvil]

Wonder what the 128GB Samsung (or a 256GB) would manage, probably more but not much I think.

[Ao1]

Whoa, over a 3 hour duration of the “normal” endurance load I’m getting a WA factor of 1.04. I’m going to leave the normal endurance workload running overnight to double check and make sure as that is way lower than what I was seeing on the last f/w.

“Normal” ASU WA

60GiB (LBA) * P/E Cycles (28) = 1680
Actual writes = 1,602

WA = 1.04

[Christopher]

Ao1,

Mine doesn't seem to be updating 177 correctly after the FW update either, so you may want to double check.

My drive is faster at the moment, but not back up to the speeds you're getting. Mine was like that before I hit the unassailable performance barrier. Now I'm happy with 130MBs. But even in the beginning, I couldn't get 160MB/s in the endurance test for very long. Basically, my average dropped to 110MBs after a while.

[Ao1]

I'm logging at 10 min intervals and the dec value is more or less changing every time th elog is updated, so it is definently working now

I'm using driver 10.8.0.1003 on SATA III btw.

[Christopher]

Hmm. Well my drive did update 177 in the beginning (though not correctly). I switched back to rst so I could SE the drive, but Ive written 500GB since hitting 14038, and its still there.


I do think that if there is one bit of advice I would give to 830 owners, it would be to over provision the drive. The Magician software will let you do it with data on the drive (I believe). The 830, especially the 64GB, needs a little extra space I would expect. Taking the 64GB from 59 down to 55 would probably work wonders for keeping the drive in shape. We (or me) may wish to try that at some point.

I'm using 10.5.1015 on an h67 SATA III. But I couldn't average 141MBs when the drive was new and with no static data. Despite the fact I was getting 160mbs writes at the loop start, my avg for the first few days was 110mbs. So really, I'm ahead of the game right now.

I do think that Samsung has tweaked some drive behavior with the new FW. I think they may have adjusted some of the GC/TRIM stuff as well. The drive acts a little differently now throughout the loop. Also, it's quite a bit faster after the SE on the new FW.

BUT...
I tested read speeds... and they're even worse now. But the write results speak for themselves.

[minpayne]

That's astonishing speed, and the WA is now close to the other models finally!

[Ao1]

Phenomenal

“Normal” ASU WA after 8 hours (and days of 4k random write abuse)

60GiB (LBA) * P/E Cycles (68) = 4080
Actual writes = 3,937

WA = 1.036

I'm now 100% convinced that 177 is the MWI and P/E cycle count.

[minpayne]

@Ao1, I wish I could press the "Thank You" button but there's no such one for this forum account yet

Now it remains to be seen how much host writes it can sustain, with the new firmware. Hopefully it will last longer than the 470, meaning that the WA is really reduced, not that they simply cheated by slowing down the MWI count without actually addressing the bug that caused unnecessary and excessive WA.

[Christopher]

New drive starting on Monday, maybe Sunday night. I'm leaving today for a short-notice trip, and won't be back till then.

But, this 'new' drive has similar seq write speeds to the 830... Almost identical. And it's 64GB too...

I've put it in my laptop to shake it down whilst I'm away... I want to be sure that it's five by five before testing begins. But there have been some initial signs that the drive has some... quirks.

In any event, I think I'll test this drive and the Samsung while waiting for the new drives to launch.

[Ao1]

Result! I had delivery issues when I ordered my 830. Out of the blue the retailer (Dabs) just emailed to say they would send me another 830 FOC to compensate! I’ve been so impressed with the 830 I’m going to use it for my OS drive when it arrives.

Anyways, a mini update

Enterprise ASU WA after 6 ½ hours

60GiB (LBA) * P/E Cycles (51) = 3,060
Actual writes = 89
WA = 34.38
Current MB/s: 3.98MB/s
MWI: 74
P/E Cycles 925

Looks like I’ve lost 0.5MB/s on the enterprise workload after the f/w update, but again this is phenomenal (and consistent) performance considering this is about the worst workload you could ever throw at a drive. (Also taking into account that the workload prevents TRIM from functioning).

[johnw]

Enterprise ASU WA after 6 ½ hours

60GiB (LBA) * P/E Cycles (51) = 3,060
Actual writes = 89
WA = 34.38
Current MB/s: 3.98MB/s
MWI: 74
P/E Cycles 925

It is interesting that whether you run the standard ASU workload or the enterprise, the 830 goes through erase cycles at nearly the same rate (cycles per hour). So the time to exhaust MWI is nearly the same with either workload.

[Ao1]

^ I hadn’t got round to looking at that in detail, but I had picked up that the wear rate was similar. A rough and ready way to look at it

4 MB/s * 34.38 = 137.52 MB/s
141.09 MB/s * 1.036 = 146.17 MB/s

According to that way of looking at it 4K is slower. I will work it out properly at some stage.

EDIT: I'll post the log file when I get to 3K