SSD Write Endurance 25nm Vs 34nm

[Christopher]

B.A.T and I started the M4 vs. 830 contest at virtually the same time. While his drive had fallen behind for a few days over Christmas, mine lost time writing for almost three days.

At this point I would have to say there is no way the 830 makes it past 500TB -- but we know the M4 can, just at half-speed.

The 830 has nearly 18,000PE cycles on it, perhaps even a few more. Johnw's 470 had a wear leveling count of 39,000ish, so I doubt the 27nm Toggle will ever hit those numbers -- if it could, it would be a lot of TBW when it finally died as write amplification is pretty good now. It went from using 544 a day for the first 10 days, to using 370 a day, down to under 200 PE a day now for over 11110GB of writes.

I think it can only make it to around 23 000 PE cycles.

[bulanula]

Thanks for the pictures and screenshots of the SF death.

What nm are those NAND chips ? 25nm ?

[Christopher]

Thanks for the pictures and screenshots of the SF death.

What nm are those NAND chips ? 25nm ?

Those are 25nm Micron *AAA Asynchronous NAND. The *AAB units are Synchronous.

Those are some great pictures though. Anvil is a man of many talents.

[Christopher]

Samsung 830 64GB Update, Day 57
FW:CXM02B1Q

TiB 399.7556
GiB 409349.72
138.11 MB/s Avg
PE Cycles:
16672

Reallocated Sectors
49152
24 Blocks, holding steady

1356 hours




OCZ Vertex Turbo 64 Update Day 13

TiB 81.1199
GiB 83066.81
82.87 MBs Avg

Avg Erase Count
1491
MWI 71 down from 73


1 Program Fail
28 Erase Fail (up from 26)
0 Read Fail
296 Hours




-----------------------------------------------------------------------

[Anvil]

Will be updating later today...

I still haven't decided on what drive to start testing, I've got 3 options, Intel, Crucial or another Force 3.
Either way it will be over-provisioned.

[Ao1]

Anvil it seem that the SF drives are really not benefiting from the 46% compression setting. As an example the MWI exhaustion for the Chronos was worse than the M225 and only marginally better than the M4. The F3 did really well but that was a 120GB drive so it’s hard to compare with the other drives being tested. For the SF drives using 46% how did the WA come out?

Also it seems that the ability to write past MWI has more to do with the NAND rather than the controller. The NAND that Micron and Intel are using seems to be in a league of its own. Unlike other Intel models I notice that the 520 is using exactly the same IMF NAND that is being used by other SF vendors.

[Christopher]

Anvil it seem that the SF drives are really not benefiting from the 46% compression setting. As an example the MWI exhaustion for the Chronos was worse than the M225 and only marginally better than the M4. The F3 did really well but that was a 120GB drive so it’s hard to compare with the other drives being tested. For the SF drives using 46% how did the WA come out?

Also it seems that the ability to write past MWI has more to do with the NAND rather than the controller. The NAND that Micron and Intel are using seems to be in a league of its own. Unlike other Intel models I notice that the 520 is using exactly the same IMF NAND that is being used by other SF vendors.

I think this is true, but the NAND choice largely decides write speed. The Plextor M2P and Corsair Performance Pros have the same write speeds as the Samsung 830 @ 128. Regardless of the controller, as most drives are able to handle the sequential writes with low WA, you would expect similar NAND drives at similar speeds to do about the same TBW.

That said, I think there is something about the SFs that results in lower TBW figures. Something inherent that means that the drive will peter out before any other controller with the same NAND would. This is why I'd like a 64GB Toshiba Toggle drive with the Marvel controller -- I bet it would last longer than the Chronos D, which made it to around half of what the F3 did. I also believe the Chronos D, with it's 32gbit devices, had RAISE like the 120+GB 64gbit drives. So the F3 and CDx were on a level playing field, mostly.

It could also be possible that Samsung's NAND is every bit as good or better than the Intel/Micron secret-NAND-stash. I would certainly hope they'd keep back the best stuff for their own drives.

[johnw]

There is a 64GB Plextor M3 that has just been released (24nm Toshiba toggle-flash). The only place I've seen it for sale is newegg, and, as is their wont, newegg is price gouging it at the moment. Hopefully after the SSD achieves wider circulation, it should come down to $99 or so.

http://www.goplextor.com/asia/index....x-64m3?start=1

[Christopher]

I've found some P2Ms, and they're less expensive than they were, but still pretty high. The problem with the 2xnm Toggle is that it's in 64gbit devices while the 32nm Toshiba Toggle and 27nm Samsung are 32gbit. I'll check it out.

I got the new MTRON MOBI 32... but it's not really any faster than the 16. I was hoping it was twice as fast, but it's not any faster in ASU. Which means it would take a lot, lot longer.



As sad as it is, the 4K IOPS are better than manufacturer rated. It's only rated at 130 4K W IOPS.

[bulanula]

What about SLC ? Or is that dead from long ago ?

Any reasonable priced X25-E successor with 34nm SLC NAND ?

[Christopher]

What about SLC ? Or is that dead from long ago ?

Any reasonable priced X25-E successor with 34nm SLC NAND ?

There is the 311 20GB Intel SLC, the upcoming Intel 711 series -- the E successor with 34nm SLC, but that's it. I really wish I hadn't accidentally killed the MOBI 16GB SLC.

The 32GB MOBI SLC is actually a little slower in endurance testing, so that makes it a non-starter. I'm efforting to get another 16GB SLC drive.

In terms of price, the X25-E's successor will be around $10 a GB. There aren't really any SLC drives for consumers these days, and it's a shame. 25nm SLC shouldn't be nearly as expensive, but it's intended market segments aren't as price-conscious. If Intel could just make a 40GB 311 SLC for under $175 US, I think many would choose the speed and endurance over cheaper, more spacious MLC drives.

[Anvil]

Kingston SSDNow 40GB (X25-V)

675.33TB Host writes
Reallocated sectors : 05 21
Available Reserved Space : E8 99
POH 6087
MD5 OK

32.79MiB/s on avg (~99 hours)

--

I'll got a few more days to decide on what to select as the next drive. (until the X25-V is having a retention test)

I agree that the F3 should have done better, I had sort of expected it to do closer to 2PiB based on it being a 120GB drive but we might be seeing the limitations of this particular NAND. (3K P/E)
Anyways ~1PB is a lot

@Ao1, Host writes were 1015TiB and NAND writes were 780-790TiB.

[Christopher]

Samsung 830 64GB Update, Day 57
FW:CXM02B1Q

TiB 399.7556
GiB 409349.72
138.11 MB/s Avg
PE Cycles:
16672

Reallocated Sectors
49152
24 Blocks, holding steady

1356 hours




OCZ Vertex Turbo 64 Update Day 13

TiB 81.1199
GiB 83066.81
82.87 MBs Avg

Avg Erase Count
1491
MWI 71 down from 73


1 Program Fail
28 Erase Fail (up from 26)
0 Read Fail
296 Hours




-----------------------------------------------------------------------

[bulanula]

Any news of that Intel 520 ? It seems it is just rumors as of now.

[Christopher]

Any news of that Intel 520 ? It seems it is just rumors as of now.

The NDA hasn't lifted yet, but they're available at retail. I've already canceled my order...

[Christopher]

Samsung 830 64GB Update, Day 58
FW:CXM02B1Q

TiB 408.6758
GiB 418484.04
144.85 MB/s Avg
PE Cycles:
17035

Reallocated Sectors
49152
24 Blocks

1374 hours



-----------------------------------------------------------------------

[Anvil]

What's wrong with the Turbo?

[Christopher]

What's wrong with the Turbo?

Nothing, I hope

I'll post it later. I was upgrading ASU from Beta 9 to Beta 11, so I just did the Samsung.

[Christopher]

OCZ Vertex Turbo 64 Update Day 14

TiB 86.7804
GiB 88863.11
82.87 MBs Avg

Avg Erase Count
1595
MWI 69 down from 71


1 Program Fail
29 Erase Fail (up from 28)
0 Read Fail
316 Hours




-----------------------------------------------------------------------

[bluestang]

Those erase fails are going quick. You have to pop that Turbo open when it dies.

[Christopher]

Those erase fails are going quick. You have to pop that Turbo open when it dies.

Yes, yes they are.

But all of the factory bad blocks were laundered, and thrown back into the good pool, so really I think of it as only 11 erase fails... Based on similar drives, like my Vertex LE 100, 128 Turbo that I've peeped out, it's gonna look just like the original. Which is what I'm pretty sure it is, just tarted up with some new FW and a new plastic dress.

[Christopher]

I know there is a way to determine actual WA on a SF drive, but that it's covered under strict NDA (though surely someone has figured it out by now). It's possible that the Chonos D and F3 could have much higher WA than it would appear, and perhaps its hidden away in the Smart data for anyone who knows what they're looking for to see.

Maybe there was a good reason that the original SF drives uses 28 percent OP.

[bluestang]

Yes, yes they are.

But all of the factory bad blocks were laundered, and thrown back into the good pool, so really I think of it as only 11 erase fails... Based on similar drives, like my Vertex LE 100, 128 Turbo that I've peeped out, it's gonna look just like the original. Which is what I'm pretty sure it is, just tarted up with some new FW and a new plastic dress.

Yes, forgot about the FW and what it did with the initial bad blocks. Hopefully it will go quite a distance.

I'll have to check my M225 data when I get to work tomorrow to see where I was TB wise with 11 erase fails.

[Christopher]

Yes, forgot about the FW and what it did with the initial bad blocks. Hopefully it will go quite a distance.

I'll have to check my M225 data when I get to work tomorrow to see where I was TB wise with 11 erase fails.

I wonder how it tracks with yours. I think my drive is good bit slower, but yours was run without TRIM for quite some time.

I did look at your drive's last stats and saw that your 1.6FW drive and my 1.7FW are using the same number of PE cycles vs. GB written, just a bit over 56GB/PE cycle.

[Christopher]

The 830 is going down for a retention test as of 10:53PM GMT -6 31 Jan 2012.