SSD Write Endurance 25nm Vs 34nm

[johnw]

Works out to 109.4MiB/s if 12hrs = exactly 12hrs. Also includes all the breaks/pauses in the program, not sure if they should be included in the average or not

Here are the period average speeds and the cumulative average (since test begun) write speeds. Note that there were 5 hours on the SSD when I started running Anvil's app.

hours	TiB	MB/s	Cum MB/s
5	0		
29	8.589	109.3	109.3
54	17.586	109.9	109.6
79	26.387	107.5	108.9
91	30.895	114.7	109.7
103	35.306	112.3	110.0

[B.A.T]

8,7082 TiB, 33 hours, Wear Leveling Count and Percentage of the rated lifetime used has gone from 97 to 96.
Avg speed for all 33 hours is roughly 77 MiB/S

[Vapor]

Thanks everyone for the write speed input

I've got:
the m4 at 77MB/s (based on data)
the 470 at 110MB/s (based on data)
the 320 at 41.8MB/s (based on data)
the V2-40 at 45MB/s (user guess)
the X25-V at 33.5MB/s (my guess)

B.A.T, can you check back on your SMART values in an hour-ish? I have a theory that the normalized value is just 100 - (raw / 30). Based on that, when the raw value hits 150 (and the raw is increasing at ~4 increments/hr), the normalized value will hit 95.

Seems the Samsung 470 has the same normalization algorithm, just with 50 raw increments per 1 normalized increment (I have been recording the raw value, your extra keystrokes are not in vain, johnw).






Am considering a normalized write-day set of charts but I am worried about over-charting and the fact that write speeds depend on SSD size, which does negate a normalized chart's ability to be used for extrapolation.

[B.A.T]

It just did. 8,930 TiB. That's good work

[bulanula]

Really nice work guys keep it up!

[johnw]

35.306 TiB, 103 hours, sa177: 41/41/2970

39.859 TiB, 116 hours, sa177: 35/35/3282

[Anvil]

I decided on using a 4.78GB VM partition for the MD5 test and it worked out just fine, meaning that the checksum compared.

I also monitored a few loops and the speed looks to have increased a bit since adding that 5 second pause.
It starts off at almost full speed (38MB/s) and then increases in speed during the next 90 seconds where it peaks at ~40MB/s.
During the last 5GB of the loop, speed drops gradually so it's basically no change from what I've been observing in the early stages of the test.

Avg MB/s for the current session is now working, it looks to have stabilized at ~35.5MB/s over the last 10 hours.

--

127.98 TB Host writes
MWI 30

[bulanula]

Getting closer and closer to that magical 0 MWI value!

[B.A.T]

48 hours, 13,2024 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 95 to 93.
Avg speed for all 48 hours is roughly 80,1 MiB

[sergiu]

I see in this thread huge differences at sequential write speed between similar SSDs like Samsung 470 vs Intel X25-V. Is this entirely related to different flash/channel configurations, different controllers & firmware optimizations? For me it looks more like software limited write speed.

[johnw]

I see in this thread huge differences at sequential write speed between similar SSDs like Samsung 470 vs Intel X25-V. Is this entirely related to different flash/channel configurations, different controllers & firmware optimizations? For me it looks more like software limited write speed.

Different kind of flash, different number of channels, different number of die per channel, different firmware. Probably different write amplification, but that is hard to know for certain on the Samsung 470. Together these can make a large difference, as you see.

[bulanula]

I see in this thread huge differences at sequential write speed between similar SSDs like Samsung 470 vs Intel X25-V. Is this entirely related to different flash/channel configurations, different controllers & firmware optimizations? For me it looks more like software limited write speed.

Intel always has limited the write speed on its sees afaik

[johnw]

Intel always has limited the write speed on its sees afaik

Intel has never throttled write speed on any of their SSDs.

[johnw]

39.859 TiB, 116 hours, sa177: 35/35/3282

44.403 TiB, 128 hours, sa177: 24/24/3805

[devsk]

44.403 TiB, 128 hours, sa177: 24/24/3805

How do I interpret 24/24/3805?

[johnw]

Read the thread.

[B.A.T]

56 hours, 15,5622 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 93 to 92.
Avg speed for all 56 hours is roughly 80,9 MiB

[devsk]

Read the thread.

Awesome explanation... 3805 value for thread is too high IMO.

[bulanula]

Awesome explanation... 3805 value for thread is too high IMO.

Lol nice one made my day! Eager to see the c300 with vapor !

[johnw]

44.403 TiB, 128 hours, sa177: 24/24/3805

48.927 TiB, 141 hours, sa177: 18/18/4140

[B.A.T]

71 hours, 20,0572 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 92 to 89.
Avg speed for all 71 hours is roughly 82,2 MiB

[One_Hertz]

172TB. 11%. Reallocated sectors still at 16.

[Ao1]

Hi B.A.T, the M4 seems to be doing rather well so far.

MWI is close to the 320, but writes speeds are much faster. The capacity of the M4 is larger, so faster write speed could be expected, but it does not seem to be having a significant impact on wear. (Compared to John's Samsung that is).

SMART attribute 181 is interesting. Looking at the literature that John posted: "This attribute tracks the number of use data accesses (both reads & writes) where LBAs are not 4KB aligned or where size is not modulus 4KB, assuming logical block size = 512B."

That seems to confirm a suspicion I had that 4KB alignment only occurs where it is possible, but in real life applications there are many reasons why 4K alignment may not be achievable.

Hi John, whoa, you will be first past the post at the rate you are going.

[B.A.T]

It's grinding away like there is no tomorrow. Top speed hangs around 100 MiB/s and it shows no sign of slowing down. I wonder when we will see the first reallocated sector?

[Vapor]

Updated charts

Raw data graphs

Writes vs. Wear:


MWI Exhaustion:


Host Writes So Far:

(bars with a border = testing stopped/completed)


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: