SSD Write Endurance 25nm Vs 34nm

[Vapor]

Finally some figures for the C300

So, 300TiB, should be great fun!

300TiB is indeed a lot

But it still won't last as long, in days, as your X25-V.

On another note, according to the "formula", a 128GB C300 with 34nm NAND should last ( 5000 * 128 ) or 640 000 GB of writes or 640 TB of writes ! The 64GB version of the C300 should last about half of that so about 320 TB etc. ???

That assumes a write amplification of 1.00x, which is a tough assumption without some sort of compression (or totally erratic performance). It does appear that the Crucials do have a nice, low WA though

[Ao1]

The Crucial will likely have 34nm nand that is spec'ed at 10K P/E cycles.

If anyone testing can confirm the nand product being used in their SSD I can try to find the manufactures specs. The V2 had Intel 34nm nand.

[johnw]

Hi John, when you were waiting for the MDF5 update did you leave the drive on idle? Maybe that period allowed static data rotation to revitalise the NAND reserve? Seems a bit strange that it is still going strong 9TB after getting to MWI 1. Even if the NAND specs are based on minimum PE cycles, that still seems quite a lot of data

I had it on for a few hours without running Anvil's app, but then I powered the computer (and SSD) down for the night, and did not power-on again until Anvil's new app was ready the next day.

I'm not surprised that it can continue going after more than 6000 erase cycles. Between possibly conservative specs for the erase cycles of the flash, and the spare-flash the controller reserves, I would not be surprised if it tops 10,000 erase cycles, or even more. But who knows? Wait for the data! We shall see...

[B.A.T]

Evening update:
177.5 hours, 51.8789 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 72 to 71.
Avg speed for all 177.5 hours is roughly is 85.12 MiB/s

[bulanula]

Really curious how much better the C300 will end up being compared to the M4

[Vapor]

Really curious how much better the C300 will end up being compared to the M4

Define 'better'

If you want the longest lasting drive, the SandForces seem like a good option

[devsk]

Define 'better'

If you want the longest lasting drive, the SandForces seem like a good option

If and when the drive decides to boot. It may ask for various limbs, certain bushman rituals and zero-swear praying! And a lot of idle time!

[johnw]

74.641 TiB, 214 hours, sa177: 1/1/6237

84.218 TiB, 240 hours, sa177: 1/1/6996

The other two unknown SMART attributes, 178 and 235, are still at 72/72/276 and 99/99/2, just as they were when the SSD was fresh out of the box.

[DragoonXX]

from the reading so far, it seems to me that the average SSD should be able to write at least 2500x its capacity (ignoring the small sample size here, where drives here are hitting ~4000+x their capacities). it makes me wonder how much a conventional consumer level HDD can handle under the same kinds of workloads.
I'd love to see what kind of per capacity comparison consumer HDDs have (from conservative estimates- a 1TB HDD should be able to write AT LEAST 2.5PB to match up with what SSDs can handle) and it seems to me that the app that's being used for testing on the SSDs would be a program of choice to test that...if anyone else is interested to see (as I'm not sure if normal HDDs have any sort of SMART attribute that monitors writes)

[B.A.T]

The Crucial will likely have 34nm nand that is spec'ed at 10K P/E cycles.

If anyone testing can confirm the nand product being used in their SSD I can try to find the manufactures specs. The V2 had Intel 34nm nand.

Here is a picture under the hood on my M4.


There are 8 of these on the pcb.

[B.A.T]

Morning update:
188 hours, 55.0254 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 71 to 69.
Avg speed for all 177.5 hours is roughly is 85.25 MiB/s

[sergiu]

I'd love to see what kind of per capacity comparison consumer HDDs have (from conservative estimates- a 1TB HDD should be able to write AT LEAST 2.5PB to match up with what SSDs can handle) and it seems to me that the app that's being used for testing on the SSDs would be a program of choice to test that...if anyone else is interested to see (as I'm not sure if normal HDDs have any sort of SMART attribute that monitors writes)

I would also like to see such a test. But unfortunately, we would not be able to do a fair compare. A 3 plates 1TB could write 2.5PiB in around 282 days (assuming 110MiB/s speed) . Now if you consider only random writes, at a blazing fast 1MiB/s speed, you get 85 years. Considering same amount of data written, there would be three possible scenarios to compare:
- sequential write (probably HDD>SSD)
- random write (SSD>HDD due to time needed to write same quantity)
- error rate and data retention
Average load scenarios are a combination of sequential/random write. What would probably make the difference will be the error rate and data retention.

[Anvil]

144.04TB Host writes
MWI 21
Reallocated sector count, still at 6.

Avg speed 33.26MiB/s (over a 15.7 hour period), MD5 6/0.

--

One can do a test of an HDD, it won't be a compare though.
It would still tell for how long (that particular) HDD would handle the stress of this test, it might go on for years or it might fail in a few weeks, as the HDD has moving parts there are other factors that could trigger a failure.

I'd say it's an interesting test to do.
I'm running a short test as we speak on an F3, looks like 8-10TiB/day is doable on an empty drive, mine is half full and avg speed is 98-99MiB/s. (one loop is set to ~12GiB of writes)

[Ao1]

Here is a picture under the hood on my M4.

There are 8 of these on the pcb.

I believe your NAND is rated for 3,000PE cycles. The OCZ Vertex 2 used Intel NAND 29F32G08AAMDB.

Looking at info I could find on the web....

MICRON 25nm 3,000 PE cycles. 4K page file. Block size 1,024K
• MT29F32G08CBACA
• MT29F64G08CEACA
• MT29F64G08CFACA
• MT29F128G08CXACA
• MT29F64G08CECCB
• MT29F64G08CFACB

MICRON 34nm 5,000 PE cycles. 4K page file. Block size 1,024K
• MT29F32G08CBABA
• MT29F64G08C[E/F]ABA
• MT29F128G08C[J/K/M]ABA
• MT29F256G08CUABA
• MT29F32G08CBABB
• MT29F32G08CBCBB
• MT29F64G08CFABB
• MT29F64G08CECBB
• MT29F128G08CJABB
• MT29F128G08C[K/M]CBB
• MT29F256G08CUCBB

MICRON 34nm 10,000 PE cycles. 4K page file. Block size 512K
• MT29F16G08MAA
• MT29F32G08QAA
• MT29F64G08TAA

Intel 34nm 5,000 PE cycles. 4K page file, Block size 1,024K
• (JS)29F32G08AAMDB
• ( JS)29F64G08CAMDB
• ( JS)29F16B08JAMDB

[alfaunits]

V2 (presuming Vertex 2?) also had 25nm versions which resulted in customer complaints over no indication of it as there was no way to differentiate the two V2 samples.
Looking at those PE cycle ratings, 200TB for a 64GB, even if 3,000 PE cycle NAND were used is extremely likely. The 3,000 are minimal or average ratings, but it should be able to handle more. Consider how much it could do fully sequential.

As for the HDD test, Google showed some stats about its SATA HDD usage, error rates and failure rates.

[One_Hertz]

190.5TB. 1%. Still 19 sectors.

I've reached the magical point except the SSD clearly doesn't care.

[Computurd]

I've reached the magical point except the SSD clearly doesn't care

great work, amazing! now to see how much longer it goes

just wanna thank you guys again for this important testing, your money, and more importantly your time, spent on this is greatly appreciated!

[bulanula]

great work, amazing! now to see how much longer it goes

just wanna thank you guys again for this important testing, your money, and more importantly your time, spent on this is greatly appreciated!

Me too. Remember, we are doing all of this in the name of science ! It is these kinds of "hands-on" tests that make XS what it is. I don't recall anything like this before. Keep it up guys ! More volunteers welcome. We are making history.

[B.A.T]

I believe your NAND is rated for 3,000PE cycles.

Thanks Ao1. Then there is no doubt that my M4 is rated for 3000 P/E cycles

Evening update:
203 hours, 59.5699TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 69 to 66.
Avg speed for all 203 hours is roughly is 85.47 MiB/s

[Vapor]

C300 Update, charts next post

21.43TiB, 93 MWI, 358 raw, 0 reallocated, 62.15MiB/s

[Vapor]

Updated charts

One_Hertz, is there a raw wear indicator for the 320? Hate to think that the only graph it'll be left participating on is just the Host Writes So Far bar graph

Host Writes So Far


(bars with a border = testing stopped/completed)


Raw data graphs

Writes vs. Wear:


MWI Exhaustion:


Writes vs. NAND Cycles:



Normalized data graphs
The SSDs are not all the same size, these charts normalize for 25GiB of onboard NAND.

Writes vs. Wear:


MWI Exhaustion:



Write-days data graphs
Not all SSDs write at the same speed, these charts factor out write speeds and look at endurance as a function of time.

Writes vs. Wear:


MWI Exhaustion:

[johnw]

One_Hertz, is there a raw wear indicator for the 320? Hate to think that the only graph it'll be left participating on is just the Host Writes So Far bar graph

Well, there is the reallocated blocks vs. TiB written. Unfortunately, the Samsung cannot participate in that (unless one of the two unknown SMART attributes turns out to track that number, but if it did, I would have expected some change in those attributes before now)

[mam72]

SSD seem really durable these days, people on other sites are right you can use a SSD like normal hard drive. They said that the drive will break before the the rewrite limit has been reached.

So does this mean doing all the stuff to limit SSD writes is a waste of time?

Also good work from all the people working on this experiment.

[devsk]

So does this mean doing all the stuff to limit SSD writes is a waste of time?

You can say that again!

[johnw]

84.218 TiB, 240 hours, sa177: 1/1/6996

The other two unknown SMART attributes, 178 and 235, are still at 72/72/276 and 99/99/2, just as they were when the SSD was fresh out of the box.

92.630 TiB, 262 hours, sa177: 1/1/7588