840 500gb at 50% OP or 840PRO at 0% OP for a server?

[Emulex]

As the pricing fluctuates the 840 500gb is going down down down while the samsung 840 pro is keeping high.

I am very happy with the samsung 830 (16 in operation) and 10 840 pro in operation now.

But this had me thinking. I can afford to run the TLC samsung 840 with 50% OP (250 out of 500gb) or I could use the 840 PRO at 256gb mode 0% OP.

Samsung 840 Pro probably has maybe 3000 P/E cycles given the die shrink? The old X25-M was 5000 P/E but die shrinks drop P/E cycles.

the TLC 840 has 1000 P/E cycles. The older SAMSUNG 830 lasted 6PB,whilst the 840 non-pro died in 1/10th of the time *source xtremesystems*.

We all know OP gives huge increases in longer running write operations (TRIM doesn't help) - But do you think a 500gb 840 @ 250gb would last longer and perform more consistently ?

The 840PRO has been OK so far, lost 1 after a week. The drive diagnosed to "nothing wrong" using the exact same controllers as the samsung 830 which have never skipped a beat. *not happy when random glitches occur*

Otherwise I might try the Neutron GTX LAMD which seem to perform better but slower. If you haven't noticed the trend. the longer lasting enterprise ssd's are hella slower than the faster consumer one. I wish you could "Throttle" the consumer drives to provide longer lifespan.

Any other folks rocking SSD in servers? A lot of good talk on webhostingtalk and lowendtalk (cheap hosting folks).

I fear I should have bought a ton more samsung 830 but alas that ship has sailed.

 

[DominionSeraph]

ITYSUTEFAFPOTMWA

caps

 

[Idontcare]

the TLC 840 has 1000 P/E cycles. The older SAMSUNG 830 lasted 6PB,whilst the 840 non-pro died in 1/10th of the time *source xtremesystems*.

That sounds about right.

I have the 256GB 830 as well as the 250GB and 500GB 840 (non-pro) and from what I read on Samsung's site the non-pro's are rated to give the end-user at least 400TB of writes over the lifetime of the SSD. You are saying one lasted for 600TB of writes on XS, I believe it.

Basically this comes down to time vs. money for you. Do some rough calculations of how long you think it will take to expend the write-lifetime of the SSD (how long for you to hit 400TB of writes to the non-pro).

Then factor in a general expectation of pricing vs time (replacement SSDs will be cheaper and cheaper as time goes forward).

If the non-pro is expected to last you 3 yrs and you just assume you will have to replace it in 3yrs at replacement cost xyz, how does that compare to buying the drive that is 2x more expensive but won't have to be replaced for 10yrs (or however the numbers work out)?

For my usage patterns I concluded the 840 non-pro was money in the bank for me. By the time I replace it I will be able to get something that is likely to cost me 10-20% the initial 840 non-pro price and it will probably be 2-3x faster to boot.

I am ok with planned obsolescence, it is only a problem when you don't plan or account for it.

 

[JellyRoll]

The tests at extreme systems did NOT take data retention into consideration. This is how NAND is spec'd, the ability of the NAND to hold data without power. The 830 died after minutes without power. It would have died around 1PB if it had been left without power for a week.

 

[Ao1]

Exactly. The theoretical PE count is specified for a reason. I’d really like to see a graph showing how data retention capability starts to degrade once the theoretical count is exceeded, but despite looking I’ve not been able to find anything the subject. I’ve got an 830 with an expired PE count that is still holding data one year on but I suspect that once the theoretical PE count is exceeded the data retention capability will drop very quickly.

 

[ArchAngel777]

The tests at extreme systems did NOT take data retention into consideration. This is how NAND is spec'd, the ability of the NAND to hold data without power. The 830 died after minutes without power. It would have died around 1PB if it had been left without power for a week.

This is an incredably useful detail that I must have glossed over.