Is SSD Over-Provisioning worth it?

[TechLarry]

Apparantly Samsung lets you change this. It's currently turned off on my 850 Evo.

Is it worth it to turn it on? The default is 10% or 46GB....

 

[evilsofa]

Do you ever fill an SSD to 100%? I never have, not anywhere close. Those who do have work flow that results in filled SSDs should use overprovisioning, but otherwise, don't worry about it. Having unused space is effectively the same as overprovisioning.

 

[TechLarry]

Works for me

 

[Nenu]

I was using around 10% on my 840 Pro.
But after reading a lot about endurance tests, they can take an incredible amount of writes such that mine will last WAY beyond 10 years without any over provisioning, so I turned it off.

 

[HammerSandwich]

Having unused space is effectively the same as overprovisioning.

Not sure I agree with this. Reserved space (through HPA or unpartitioned area) tells the SSD that it has available storage which it fully controls. Partitioned space is controlled by the OS, whether it's empty or not, requiring the SSD controller to perform garbage collection after TRIM commands come through. While TRIM clearly helps, it runs intermittently, so the SSD should be more proactive about freeing space with OP.

Regardless, I am speculating here & don't know of any test results in this area. Anand's consistency benchmarks come close, but they fill the partitioned space 100%.

 

[JoeComp]

The SSD treats TRIMmed blocks the same as overprovisioned blocks. So evilsofa is correct, assuming the filesystem is passing TRIM commands along to the SSD. On the other hand, if you are running without TRIM, then overprovisioning could be helpful.

 

[HammerSandwich]

I agree on both points.

However, TRIM does not happen in real time. Further, Windows schedules disk optimization weekly. Do you really believe this combination works exactly the same as giving the SSD controller its own 16GB or whatever to work with? I sure don't.

 

[JoeComp]

Yes, as I said, TRIMmed blocks are the same as overprovisioned blocks as far as the SSD is concerned. Your repeated statement about TRIM not happening in real time is a red herring. If you never fill your drive up past, say, 90%, then the only time it would matter how long TRIM takes would be if you had a really strange workload where you fill the SSD up to 90% then delete a lot of data, then fill up to 90% again just a few seconds later. In all common workloads, OP of 10% would be equivalent to TRIM and never filling past 90%.

 

[ToddW2]

You don't have to guess look at sites that review SSD they usually do % OP and compare to NON-OP.
Every case I've seen it is worth it.

 

[evilsofa]

You don't have to guess look at sites that review SSD they usually do % OP and compare to NON-OP.
Every case I've seen it is worth it.

Those performance consistency tests (such as this one) are done with the SSDs 100% filled with data. As I said, if you are working with 100% full or nearly full SSDs, you should OP. However, if you always have significant free space, that empty space is effectively the same as OP.

 

[michalrz]

Not sure I agree with this. Reserved space (through HPA or unpartitioned area) tells the SSD that it has available storage which it fully
controls. Partitioned space is controlled by the OS, whether it's empty or not, requiring the SSD controller to perform garbage collection after TRIM commands come through.

I thought HPA is a lower-level manipulation than editing the partition table.
As the SSD controller throws data around its cells, it would have to understand the partition table, monitor its changes, do the right calculations on the fly.

While the command 'read native max address' will still return the true upper LBA, the fake value returned by 'identify device' is what the OS adheres to.

Personally I still have that spinner mindset according to which I try to always leave 2-5% unused but partitioned. So that the OS has somewhere to write to should anything weird happen like a crash requiring dumping buffers/cores.

I don't think wear is an issue to be honest. I have an old laptop at work with XP still on it (I need to run really old stuff on it time to time) so there's no TRIM. And after many months of running with a 80% full Kingston 240GB SSD - no signs of trouble or degradation. But nevertheless I did leave like 3% unpartitioned just in case. I feel that it's the manufacturer who should be answering this question because of the proprietary nature of the controller. Stuff like deferred commands, auto-checks, utilizing compression (like some older SSDs) is unpredictable and "they" could be hiding it under the hood.