SSD Write Endurance 25nm Vs 34nm

[Christopher]

No, the WA is definitely not 1.06. The 5444 erase cycles makes sense (and 100 is definitely wrong) since the normalized value has already reached threshold (it must have done at either 3000 or 5000).

You could be right, but my gut tells me neither wear leveling count could be accurate. And I still can't account for the crazy performance drop. The drive didn't go from ~100 PE cycles to over 5000 overnight, it happened instantly. So you could be right, I just have no way of knowing.

The PE cycles havent increased all day either since hitting 1/1/5444... the threshold is 0, but maybe it's at the end of the line.

Right now it's on track to write 200MB an hour... that's not good.

[johnw]

The PE cycles havent increased all day either since hitting 1/1/5444... the threshold is 0, but maybe it's at the end of the line.

I meant that the actual value of 5444 erase cycles makes sense. The fact that it stuck around 100, then increased to 5444, and now is not increasing, does NOT make sense. That certainly seems like a firmware bug.

Combined with your weird write-speed decrease, which could be another firmware bug. I suppose it could also be some sort of throttling that comes in after the SSD depletes MWI, but my first guess is a firmware bug.

[Christopher]

I may try re flashing the fw but the Samsung software may not let me. And it was slower yesterday... The SE helped a lot, but its still nowhere close.

UPDATE
I SE'd it again. Put a partition on it, and ran CDM again:

[Anvil]

The writes are bad but where did the reads go?!

Running a 100MB test would normally be on the small side and as the Samsungs have a cache that is larger you should at least do 500MB.

I'd let it idle for a day or so, it might rely on background GC although a SE should normally take care of all such anomalies.
Could you try running HDTune Read Benchmark using 64KB and 4MB?

[Christopher]

The writes are bad but where did the reads go?!

Running a 100MB test would normally be on the small side and as the Samsungs have a cache that is larger you should at least do 500MB.

I'd let it idle for a day or so, it might rely on background GC although a SE should normally take care of all such anomalies.
Could you try running HDTune Read Benchmark using 64KB and 4MB?

The earlier post first documenting this used CDM 1000mbs. I was just in a hurry. With the many subsequent ones.

I'm film the HDTune.

[iceblade]

It really looks to me like the drive throttles down its writes because its past the MWI... Read speed should be unaffected though when testing CDM 1000mb.
I'm very curious though about how much total writes the drive can take.

[Christopher]

With respect to the CDM numbers, look at the past couple of posts with CDM shots. The first two were with 1000MB, 5 reps -- and the best numbers are being displayed. One of 5 would be much better, and the other 4 would range from low to super low. As far as the writes go, they look better than the reads, but it's not sustainable. Running the endurance loop, the beginning loop instant speeds start at 100 and drop into the 50MBs range.

Secondly, I SE'd the drive many times, in addition to multiple Magician performance optimization runs. Then, just grasping at straws, I tried one of the secure drive wipes in Acronis Drive Cleanser (the fast one that just writes zeros to every LBA). Sequential writes seemed to help at first, but didn't really do anything.

I tried running the drive with Intel RST and MSAHCI, and on two different systems.



I ran HDTune several different ways, but regardless of which benchmark I run, there is something wrong here.


64KB


4MB


I will let the drive idle the rest of the day, but I don't see why that would work but secure erasing doesn't.

[iceblade]

Then it def looks like something is wrong, read speeds should be unaffected...

[Christopher]

if you look at the above 4MB hd tune shot, the depressed reads roughly correspond to the amount of static data on the drive. However, most of the CDM shots are with no data on the drive.

[bluestang]

Man, the days just seem so boring for me now for some reason. LOL

[Christopher]

Man, the days just seem so boring for me now for some reason. LOL

Would you like to buy a Samsung 830 SSD, gently used?

[iceblade]

Christopher, your results really make me regret buying a Samsung 830 128Gb and not going for the Crucial M4. Can it be that this drive is so bad that after exhausting the MWI it drops to ridiculously slow speed for BOTH write AND read?

[Christopher]

Christopher, your results really make me regret buying a Samsung 830 128Gb and not going for the Crucial M4. Can it be that this drive is so bad that after exhausting the MWI it drops to ridiculously slow speed for BOTH write AND read?

What happened I'm not sure. But the drive is awesome.

Are you planning on writing 128TB a week? If not, you should be fine. Somehow, when I made it so more of the drive was being written to each testing loop the drive got in a bad way. But I also think it has to do with writing so much with out a reboot.

The 64GB 830 has written 90TB so far, so some drives will take it in stride while others won't.

But this drive is hard up at the moment. This is after idling for six hours

[bluestang]

Would you like to buy a Samsung 830 SSD, gently used?

No, but I'll trade you for an even more gently used M225!

[Christopher]

No, but I'll trade you for an even more gently used M225!

I actually just got a 60GB Vertex EX SLC in the mail today -- brand new, 1.3FW -- for $66... if I didn't think it would take either excruciatingly long to endurance test it, or die in an embarrassingly short time I'd do it in a heart beat... it gives your M225 Turbo a run for it's performance money.

I also have a new Imation S-class SLC coming in the mail, but with 120 4K write IOPS, it's probably nothing to get excited about. WA would go through the ceiling with that drive, and I think I could kill it pretty quick.

Either way, I'm looking past the 830 at this stage. Maybe I can get it back to health... maybe not.


EDIT

I SE'd the drive again with HDDERASE, and it made no difference.
An endurance loop with no static data (just after the SE) on the drive is just as slow as before. CDM still shows terrible results.


I have a theory... but it's not a good one yet. I'm going to continue with the testing.

[Anvil]

@Christopher

Could you try formatting the drive?

Make sure to deselect the Quick Format option as it will make a huge difference, you can stop at 10-20% and run ATTO to check if it works for you, you need to perform a quick format if you abort the full format in order to run tests of course.
(a quick format just won't do sometimes, even if it should include a TRIM)

[Christopher]

I've heard of this mystical Full Format option (yeah, I still remember how to perform a full format with Windows). I'm doing it now, but I already did something very similar.


UPDATE
Full format didn't do anything.

When using CDM 1000MB *5, the first read result is 80MBs every time. 80/320/100/270/119 is common for the sequential reads.

I think the firmware is broken, but I can't reflash it. You have to use the Mgician to manage fw updates, and I can't reflash the same version.

Until Samsung releases a new fw I'm stuck with this drive...





Unless this is some crazy wear leveling scheme, and nothing is broken at all. Perhaps this is what happens when too much wear occurs on some cells, in which case running the drive for a few days without any static data may help.

[Christopher]

Samsung 830 64GB Update, Day 12
FW:CXM01B1Q

GiB written:
96974.70

Actual Writes:
97765.9

Avg MB/s
75.96MBs down from 105MBs

PE Cycles
6090

Reallocated Sector Count
20480 (10 blocks)
8K pages, 1MiB blocks

292 hours





--------------------------------------------------------------------------------
I'm currently hoping that running the loops with no static data will help.

I think this has something to do with the wear leveling algorithms.

[B.A.T]

Finally after 3 days of christmas frenzy trying to get all the details fixed before the big day, here are the latest update

Kingston V+100
299.9157 TiB
1469 hours
Avg speed 25.37 MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
147,5687 TiB
19688 hours
Reallocated sectors : 00
MWI=15 to 0 and up to 100%
MD5 =OK
51.23 MiB/s on avg


m4
83.9249 TiB
304 hours
Avg speed 81.20 MiB/s.
AD gone from 65 to 52.
P/E 1461.
MD5 OK.
Reallocated sectors : 00
http://www.diskusjon.no/index.php?ap...tach_id=472286

[Anvil]

The Force 3 was back on power again late Friday night so it was a disconnected for ~54-56 hours, it has been idling most of the time since it was back online.

I'll make a report in a few hours.

[Christopher]

Bluestang,

I've changed my mind about Indilinx 1.7 firmware. I've been having problems with the one drive I put on 1.7, but the amount of WA I was experiencing with earlier FW is probably worth the risk.

But my 128GB Turbo running 1.6 did lose some 4K read performance, but seemed to gain more 4K write. Two Agility 60s didn't really change at all, and the random performance of the 60GB Vertex EX is so good I didn't want to risk upgrading it. If the M225 had been used in a normal (normal for other people ) way I think it would be a no-brainer. In the endurance testing loads, I think it would probably be good for just enough to get you over the 1PB hump. My Turbo 128 1.6FW has less than 120GB host writes, but the PE cycles had exploded and MWI was dropping quickly compared to the Agility 60 on 1.7 which was being used without TRIM.

[Anvil]

Kingston SSDNow 40GB (X25-V)

564.87TB Host writes
Reallocated sectors : 05 14
Available Reserved Space : E8 99
POH 5117
MD5 OK

35.92MiB/s on avg (~19 hours)

--

Corsair Force 3 120GB

01 88/50 (Raw read error rate)
05 2 (Retired Block count)
B1 47 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 544724 (Raw writes) ->532TiB
F1 725061 (Host writes) ->708TiB

MD5 OK

106.53MiB/s on avg (~19 hours)

power on hours : 2098

--

I had to restart the computer about 19 hours ago as something was off and so almost nothing had been written between Friday-night and Saturday-night. (it was just very slow)

[Christopher]

Alright, the 830 is really starting to piss me off. Something is not right with it, but as far as I can tell, speed is the only thing affected

It could just be some crazy amount of write amplification, but I don't understand why reads are affected.

Samsung 830 64GB Update, Day 13
FW:CXM01B1Q

GiB written:
103395.55

Actual Writes:
97765.9

Avg MB/s
65.83MBs

PE Cycles
6407

Reallocated Sector Count
24576 (12 blocks)
8K pages, 1MiB blocks

320 hours




I added 39GB of static data to the drive to see what would happen.

I'll try to keep the updates coming in while I'm out of town for Christmas. I won't be back for almost two weeks, so I hope the drive and system can make it without me being here physically.

[Anvil]

Kingston SSDNow 40GB (X25-V)

565.88TB Host writes
Reallocated sectors : 05 14
Available Reserved Space : E8 99
POH 5126
MD5 OK

34.86MiB/s on avg (~28 hours)

--

Corsair Force 3 120GB

01 90/50 (Raw read error rate)
05 2 (Retired Block count)
B1 47 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 547268 (Raw writes) ->534TiB
F1 728447 (Host writes) ->711TiB

MD5 OK

106.52MiB/s on avg (~28 hours)

power on hours : 2107

--

I had expected the 830 would perform more or less like the 470, the one thing I expected was that avg writes would increase and not the other way around.

[Christopher]

Anvil,

There is something wrong with it. I though about RMa'ing it and starting over, but I was hoping... something... would happen to 'fix' it. Its okay. I have a new drive to test come January.