SSD Write Endurance 25nm Vs 34nm

[B.A.T]

m4 update:

354.9926 TiB
1167 hours
Avg speed 89.60 MiB/s.
AD gone from 164 to 149.
P/E 6221.
MD5 OK.
Still no reallocated sectors



The Kingston V+ 100 will be ready in about 2 hours. I just need to set it up first.

[Vapor]

m4 is the 2nd drive past the 2:1 ratio (actual lifetime : MWI lifetime).

Still no reallocated sectors/blocks and it's using 'lesser' 25nm NAND. Very impressive.

[B.A.T]

And now:
The Kingston V+100 64GB


CDI



Some benchmark:

AS SSD


AnvilPro


ATTO


A picture of the nand inside



The Kingston is filled with the same content of the m4 (just copied from the M4)

[Vapor]

Did any SMART values move after the benchmarking/static data writes? Doesn't look like there's many SMART attributes worth tracking with how they read in the first screenshot.

[B.A.T]

I haven't written more than 55 GB yet but I'll know soon.

[Computurd]

thanks for the Kingston! that is the value leader, so many will be interested in that one. you guys are awesome with your contributions!
poor vapor, his graphs are getting more complex and bigger!

[Ao1]

I'm looking forward to seeing what the Kingston can do as well. Anand stated it had rock solid sequential performance, but overly aggressive garbage collection. Guess we will soon see.

That NAND part # is : TH58NVG6D7FBAK2 (?)

[B.A.T]

That nand part# looks right. The picture was a little blury.

I think the GC is messing up the seq speed. The avg at the moment is ~79 MiB/s

[Christopher]

How can the M4 have so many Non 4K Aligned accesses and have such low write amplification?

Most of that would be writes anyway given the ratio of reads to writes.

I was trying to figure out the write amplification of my old Agility60 with 10000PE rated NAND -- I've only written ~1400GB to it, but it's down to 84% MWI with an average PE of ~1600. I started keeping track of it a month ago after seeing how low the WA was on the M225 > Vertex Turbo, even without trim.

[B.A.T]

Did any SMART values move after the benchmarking/static data writes? Doesn't look like there's many SMART attributes worth tracking with how they read in the first screenshot.

AD went down from 200 to 199 after little over 0.65 TiB

[Vapor]

Looks like AD is behaving similar-ish to MWI. If AD 199 = 99.5 MWI and your NAND is rated for 5000 cycles, then you've used 25+ cycles to write .7TiB (roughly accounting for static writes + benchmarks), then you're looking at ~138TiB MWI Endurance and a WA of ~2.25x. That all seems plausible, especially so early into the process

[Vapor]

C300 Update

246.07TiB host writes, 17 MWI, 4158 raw wear indicator, 2048/1 reallocations, 63.85MiB/sec (13hr average after a restart, IDK why it's suddenly so much higher than it was the rest of the test), MD5 OK


SF-1200 nLTT Update (from much earlier today...MWI has ticked down a notch in awhile)

115.813TiB host writes, 78.719TiB NAND writes, 63 MWI, 1259.5 raw wear (equiv), wear range delta 3 (still), 56.5MiB/sec, MD5 OK


C300 and m4 are just a few days from out-living the Samsung 470 in terms of days and have just 1 combined reallocation (and the C300 won't even have exhausted its MWI when it passes the 470).

[B.A.T]

m4 update:

357.7790 TiB
1177 hours
Avg speed 90.51 MiB/s.
AD gone from 149 to 147.
P/E 6270.
MD5 OK.
Still no reallocated sectors

The m4 was not running the for 2 hours when I made the Kingston ready.


Kingston V+100:

2.2175 TiB
9 hours
Avg speed 78.97 MiB/s.
AD gone from 199 to 194.
P/E ?
MD5 OK.
No reallocated sectors

[devsk]

C300 and m4 are just a few days from out-living the Samsung 470 in terms of days and have just 1 combined reallocation (and the C300 won't even have exhausted its MWI when it passes the 470).

But Samsung 470 is 20% faster than M4 in writing and got there first....

[johnw]

Kingston V+100:

2.2175 TiB
1177 hours
Avg speed 78.97 MiB/s.

I was not expecting the Kingston V+100 to be slower than the Samsung 470, let alone also slower than the m4. That is disappointing.

[B.A.T]

I agree with you. The Kingston V+100 can only sustain the adverticed speed in bursts (like AS SSD and ASU) but not under stress like the endurance test.

[Ao1]

I'm surprised by the write speed. Anvil's app is mostly sequential. The Anantech review stated that even with a highly fragmenting work load write speeds would not drop.

"The SSDNow V+100 has wild swings in power consumption during our random write test ranging from 1.25W to 3.40W. The swings would happen several times in a window of a couple of seconds. The V+100 is aggressively tries to reorganize writes and recycle bad blocks, more aggressively than we've seen from any other SSD.

The benefit of this is you get peak performance out of the drive regardless of how much you use it, which is perfect for an OS without TRIM support - ahem, OS X. Now you can see why Apple chose this controller.

There is a downside however: write amplification. For every 4KB we randomly write to a location on the drive, the actual amount of data written is much, much greater. It's the cost of constantly cleaning/reorganizing the drive for performance."
http://www.anandtech.com/show/4010/k...lus-100-review

EDIT: Power consumption would be an interesting way to evaluate NAND wear.

[[XC] Oj101]

Is there any way to get CrystalDiskInfo or Intel's SSD Toolkit working with a single drive in RAID mode? The controller has an array as well, so the SSD has to be in RAID mode as well.

Another thing, what exactly is WA?

Loving the thread guys

[Ao1]

WA = Write Applification

[bluestang]

Is it me or does it look like CrystalDiskInfo is seeing the Kingston V+100 like a regular spinner HDD?

[bluestang]

M225->Vertex Turbo 64GB Update:

130.0 TiB
503.6 hours
MWI 33 (drops by 1 for every 50 raw wear)
3383 Raw Wear
110.19 MB/s avg for the last 15.8 hours (on W7 x64)
MD5 OK
C4-Erase Failure Block Count(Realloc Sectors) at 2 (Bank 6/Block 2406; Bank 3/Block 3925)

[Ao1]

It does appear to show that NCQ is not enabled. The AS SSD bench also seems to show that NCQ is disabled. Can't see how that would impact the sequential write speed though on Anvil's app. NCQ does help reduce WA however.

[johnw]

It does appear to show that NCQ is not enabled. The AS SSD bench also seems to show that NCQ is disabled. Can't see how that would impact the sequential write speed though on Anvil's app. NCQ does help reduce WA however.

Not all SSD firmwares and controllers implement NCQ for writes, and for those that do, not all implement it well. The Samsung 470, the Kingston V+100, and the Intel 510 all perform relatively poorly for high QD writes. Interestingly, those are also the SSDs that have among the highest sequential write speeds.

[Anvil]

Not all SSD firmwares and controllers implement NCQ for writes, and for those that do, not all implement it well...

You can say that again
I'm performing a few small tests on an WD SLC drive just to see how bad performance can get during the Endurance test.

It looks like a terrible performer when mixed block sizes are being continuously written to the drive, a small break brings speed back up again but it doesn't last for long.
Will post a few screenshots when I've gotten a few more sample rates.

[bulanula]

It's interesting to see the difference TRIM has made to bluestang. Significantly faster write speeds and reduced WA.

Nice to see that a realistic scenario is being tested on the Indilinx drive. Also very eager to see if the Samsung 470 will retain that data for much longer without MD5 errors etc.