Sandforce Life Time Throttling

[Ao1]

Its running great. 64.23MB/s to be exact.

Anytime now E6 is going to drop to 90.......

[Ao1]

E6 100
E7 99%
E9 11,078 GB

E6 raw values
0 81 100 0 0 0 100 (1byte)

The relationship between E9 and the raw value of E6 is linear. (The graph from #299 remains the same).

AVG 64.53MB/s

SPECULATION

E6, rather than E7 is based on PE cycles. When E6 drops below a threshold write speeds will be throttled and E7 will eventually level out to E6.

Without throttling (and assuming E6 is PE) the theoretical PE cycle count would run out in 148 hours at its current rate.

So, if E6 is below E7, throttling is likely to kick in at some stage. If E6 is above E7 throttling is not going to be a problem.

Based on the ~35,000GB write capacity projected by E6 and a three year warranty (26,280 hours) you could write 1.33GB per power on hour without being throttled, which is well in excess of normal client use. On the other hand heavy writing with low power on hours could be problematic. 1.33GB assumes the SSD is powered on for the entire 26,280 hours.

[Anvil]

I'm quite sure they are operating at another scale than the 24h clock, most likely 8h or less, anyways, it's not likely to be bitten by the LTT issue.

Some of the advice for the SF1 series controller that is given on forums indicates otherwise (the "let it idle advice") is given based on GC activity, lets hope that this isn't needed on the SF2 series drives.

So, in a few days we should know if it works like the V2

[Ao1]

As the MWI has dropped to 99% it would be reasonable to assume that the credit period has expired if it was based on PE cycles. The endurance app is however still happily plodding along. I'm going to guess that it will do so for the same 7 day duration that occurred with the V2. That would confirm that the credit period is based on time not PE cycles, BUT if it runs for 7 days (168 hours) at the current speed the theoretical PE cycle count will expire if E6 is correct.

Time will tell

[Timur]

Will you check the compression ratio of 0fills, too?!

[Anvil]

Talking about compression.

Did anyone notice that the Indilinx Everest is "optimized" for compressed files?

Link to press release

I created a new thread for it

[Computurd]

so another manufacturer jumps on compression! interesting....

[johnw]

Talking about compression.

Did anyone notice that the Indilinx Everest is "optimized" for compressed files?

I read that as meaning that there is little or no performance penalty when writing compressed files to the SSD. Whether the controller is compressing the data is another matter.

[Vapor]

Will you check the compression ratio of 0fills, too?!

I thought he already did

so another manufacturer jumps on compression! interesting....

Other way around, I think...if it's optimized for compressed files then it's optimized for files SandForce 'bottoms out' on, i.e., incompressible (which sounds like every non-SF controller out there, equal performance for all manners of compressibility).

[Timur]

Sorry, missed that part. Thanks Ao1!

Looks like more than 10%, but worse than what I would have thought. Do you think compression improved, or did just measurement improve?

[Ao1]

E6 100
E7 98%
E9 13,269 GB

E6 raw values
0 74 100 0 0 0 100 (1byte)

AVG 64.68MB/s
Power on hours 56

[Computurd]

Other way around, I think...if it's optimized for compressed files then it's optimized for files SandForce 'bottoms out' on, i.e., incompressible (which sounds like every non-SF controller out there, equal performance for all manners of compressibility).

oh i see. thanks. I misinterpreted that LOL

wonder if thats just shameless marketing or if there will be an appreciable boost in performance with that type of usage v a standard non-compressed SSD. i guess this actually makes it more interesting :P

[Ao1]

E6 100
E7 97%
E9 16,602 GB

E6 raw values
0 63 100 0 0 0 100 (1byte)

The relationship between E9 and the raw value of E6 remains linear.

AVG 64.48MB/s

[Ao1]

E6 100
E7 96%
E9 19,619 GB

E6 raw values
0 52 100 0 0 0 100 (1byte)

AVG 64.27MB/s

Power on hours 83 hours

Just coming up to the half way mark to 7 days, which (within an hour) is when the V2 started to throttle writes.

The relationship between E9 and the raw value of E6 remains linear. At this rate the E6 raw value will expire before 7 days.

[Ao1]

E6 100
E7 95%
E9 23,103 GB
E6 raw values
0 40 100 0 0 0 100 (1byte)
AVG 64.27MB/s
Power on hours 98 hours

EDIT: To explain how I got some of the figures in the middle graph. The MWI dropped to 99% at 11,078GB. I averaged out subsequent drops in the MWI vs writes to establish a theoretical figure for 100%, which came to 8,592GB.

MWI at 95% and 23,103GB
- Exclude 8,592GB and 5% = 14,511GB
100% = 290,220 GB (+8,592GB)
The life curve is projecting 35,000GB

There is an approximate scale factor of 8.5 between the MWI projection and the Life Curve Projection. (Which is remarkably close to the write speed reduction factor that occurred on the V2 ).

[Ao1]

E6 100
E7 94%
E9 27,748GB
E6 raw values
0 24 100 0 0 0 100 (1byte)
AVG 63.99MB/s
Power on hours 118 hours

EDIT: I've added a summary of the V3 in the first post. I will only update charts in the 1st post from now on.

[Anvil]

Already past 30GiB, getting closer

[Ao1]

E6 100
E7 92%
E9 33,322GB
E6 raw values
0 5 100 0 0 0 100 (1byte)
AVG 63.88MB/s
Power on hours 142 hours

[Ao1]

E6 90
E7 92%
E9 35,097GB
E6 raw values
0 0 90 0 0 0 90 (1byte)
Power on hours 148 hours

Write speeds are dropping quickly.

[Anvil]

How quickly?

148 hours is 6 days 4 hours and you have just made it past 35TiB.

[Ao1]

I'm not sure if it is fully throttled yet. I just stopped the app to run AS SSD. Screenshow below.

When the raw 1byte E6 value got to the equivalent of -1 the "normal" value dropped to 90. As soon as that happened write speeds started to drop.

It seems that the 1 byte value of E6 was the threshold for throttling to kick in, rather than a fixed time duration.

The V3 could write faster than the V2 so it took less than 7 days. The faster you write the faster the raw 1byte E6 drops. Not just speed either, but the type of writes. 4K full span would have throttled the drive quicker than sequential.

AVG on the app is currently 21.27MB/s.

[Ao1]

24 hours later running Anvils app. Now I will leave the drive to idle for 24hours.

[Computurd]

wow! man that is big time throttling!

so... have you made any calculations as to how many GB per hour it took to throttle it yet?

[Ao1]

^ I posted the Excel file in the first post that records everything I observed. (I'll tidy up the first post later)

• 35,097GB/ 149 hours = 235GB per hour
• Out of the 35,097GB, 8,592GB was within the credit zone.
• Avg amount of writes per drop in the E6 1byte raw value = 277GB
• Disparity factor between MWI projection and the Life Curve Projection = x 8.5

The drive is being left to idle for 24 hours. I will then use John's method to calculate how much data can be written per day.

I'm also monitoring E6 to see if it recovers will idle time. (Not sure if it will recover in 24 hours, but we will soon see.)

When this is done I will look much deeper into compression factors.

[Computurd]

Kudos on your findings, great work!