SSD Thread (Benchmarks, Brands, News, and Advice)

It’s a great drive for most home users and gamers at the cost. The Corsair MP510 is also a good one , it uses the same NAND and controller and is similarly priced normally and is factory over provisioned (12-13%) a little more to keep up full drive performance and latency hence its advertised as a 960GB or 1920 GB SSD.


The Samsungs are only better if you need sustained performance especially after the SLC cache is depleted and also when it comes to full (fully occupied by data) drive performance where the better controller can keep latency down and keep performance up compared to competitors when at factory OP levels (6.8%). This is especially true for the PRO drives. If you OP cheaper drives to at least 12-15% like the MP510 does by factory default that last issue should mostly be mitigated.

As for 30% OP on top of the 6.8% factory default, for most home users and gamers its a bit excessive unless you have some niche needs I’d say leave it at default as you won’t notice a difference in normal use and let the controller use any unused space as OP, I personally usually partition my drives 1TB down to 800 GB and 2TB drives to 1.6TB. So 22% if you include the 6.8-7.0% factory OP which you can’t see/touch.

With drives though with weaker Silicon Motion controllers like the HP EX950 you will see a doubling of latency and I believe performance drops to like 1/4 when drive is nearly filled if you don’t have like at least an additional 15% OP over the factory default (6.8%).

NVME Drivers
As for drivers I’ve seen quite a few tests and between nvstore (Microsoft default), third party (ie Samsung) and IRST there were negligible differences in most cases. I have the Samsung driver installed with my 970 Pro but it doesn’t really change much in tests. Additionally with third party drivers sometimes they can be bad like Samsung’s nvme driver last year where it caused BSODs and other issues with Hyper-V I believe. So stick to default msahci with your bios set to AHCI. Use the IRST drivers if you have RAID setup.

 

My Sabrent 1TB drive is 954GB usable, so I guess it's OP'd by 4.6% from the factory. I set aside an additional 287.57GB as unallocated - so that equates to a total of 666.28GB usable. If you say it's a 1TB drive then that's 34% overprovisioned, but I was basing it off the 954GB max usable size, so that would equate to a 30% OP - I'm not sure which is true/real, but either way I don't need all the space on the drive at the moment so I may as well OP it by at least as much as a I have already. I don't have any niche needs that you mentioned, and if I need the space then I'll just reduce the OP, but now I've got 3 NAND type drives in my PC right now and those SATA SSD's are virtually empty now I've transferred almost everything over the NVMe drive - so I'd probably just start filling the SATA SSD's rather than significantly reducing the OP on the NVMe drive. I got 250GB of free space left on the NVMe before I would even need to start considering decreasing the OP on that drive.

Yep, I'm using AHCI mode, and the default Microsoft drivers, so I guess that's a good combination like you said at the end there - I'm not gonna use RAID.

 

Just some clarification on factory OP: 1TB drives have 1024 GB of actual NAND flash so 954 GB of user available space means you have ~6.8% factory OP that you can’t touch. Any OP you add is in addition to that. So on a drive advertising 960 GB like the MP510 with 893 GB available user capacity it’s got 12.8% OP as the drive is still a 1024 GB unit in reality. Some write heavy enterprise drives advertise 800GB (still 1024 GB of Flash) with like 745 GB of user space so they have like 27% factory OP.

 

Also add MiB or MB calculations that different OS uses.
Is OP still relevant on small drives? I did it and ran out of space and had to merge the reserved OP back to primary partition.

 

The (Gibibytes) to (Gigabytes) conversion difference is used as the base 6.8% OP on most drives as the drive actually has space in Gibibytes (Windows reports space in Gibibytes) but drive manufacturers advertise user available space in Gigabytes. Some Linux distros like Mint and also MacOS report in Gigabytes that’s why it might seem like you are getting the full 1TB in those OSs.

1024 Gibibytes of Flash (the actual total space on drives) equates to ~1100 Gigabytes of user available space if there was no factory OP whatsoever.

So with factory OP at 6.8% (76 Gigabytes = 70 Gibibytes) of NAND Flash you get 1024 Gigabytes = 954 Gibibytes of user available space. Some drives advertise 1000 Gigabytes which is equal to 931 Gibibytes of user available space for those, so they have even more OP (9%) (100 Gigabytes = 93 Gibibytes)

So if you put additional 30% OP on top of factory default you actually have around ~37% (on a 1024 GB drive) and 39% (on a 1000 GB drive) of space as real OP on the drive.

 

Ah, in that case I've got a 35% OP on my Sabrent NVMe drive then: 666GB usable / 1024GB = 0.65 => 65% available = 35% OP. I'll report back with any more info about my NVMe drive: if it fails, or if there's any interesting performance notes I learn about it in the future - either disappointing or pleasing. Very pleased with the cost per GB and the performance so far, it's literally the same cost per GB as a 1TB SATA SSD, but with NVMe 970 Evo Plus performance - good deal!

 

When I tested OP'ing so many years ago, I too thought at first that there was a specific and absolute number that the magic began. I quickly learned that was false. Identical nand, identical controllers and seemingly identical RAM and layout meant little, because of what each model was tweaked and designed for within the firmware.

That led me to ignore any OP'd amounts and just concentrate on the OP'ing I was able to do. 33% is still the minimum I recommend and use (up to 65% for SSD's used as scratch/temp drives, continuously).

The OP'ing a manufacturer does is simply to keep warranty calls at bay, not a way to give you maximum performance over time.

The OP'ing I still recommend doing gives me the performance the drive is capable of in a consistent fashion while also giving me the longest nand life possible too.

As I've stated before, whether the drive is at 48% 'true' OP'ing or really only 33% OP'ing, this absolute number doesn't matter. What matters is the availability of the power/performance the SSD can offer (up to its rated maximums) on an effectively sustained basis.

Anyone buying and using anything other than Optane based SSD's and not OP'ing the drives by at least 30% is not getting the maximum performance they paid for.

Their choice, but with the ease of being able to expand the working partition if the capacity is really needed, this step in a great platform setup is a no-brainer, and is performed on all my systems including any I am asked to set up too.

In the case of having to reduce the OP'ing by expanding the working partition? Simply indicates that the current workflow and/or SSD capacity choice needs to be changed, where possible.

 

Sounds like I'm at about the right level of OP for my drive then.

Optane doesn't need OP'ing then? Is Optane the next logical step in terms of SSD mainstream adoption in the future or are there other techs that are more likely you think?

 

Optane drives have OP by factory default just like regular NAND flash based drives. Their advantage is in 4K random read/writes which is pretty important and they have lower latency. Right now cost and power draw are what holding them back but I’m sure over time they may be able to overcome that but they won’t be replacing SSDs in the general consumer market anytime soon.

I don’t completely disagree with tiller but I do refrain from making people put such large OPs without context, my personal view is the following:

As for OPing it’s really very workload dependent, for an “average” home user / gamer who uses their computer mostly for web surfing, office apps and games 30-40% might be a bit excessive unless they care so much benchmarks/peak speeds, because in real world use such as boot times and load times they won’t notice much difference.

However if you do a lot of write intensive stuff like video exports and other intensive stuff regularly, yeah you want at least like 12-15% ish minimum (From tests I’ve done on 970 PRO/EVO) on the Phison E12 based (Sabrent & Corsair MP510) and Samsung Phoenix based drives like the 970 PRO for good speeds. You definitely want a much much higher minimum OP on weak Silicon Motion controller based drives in such cases as drives like the HP EX950 which have their speeds drop to like 1/4 and latency spikes to like 4x when the drive is nearly full. On TLC drives the dynamic SLC cache is reduced based on how full the drive is so for longer sustained writes you definitely would want more OP.

 

@Robbo99999 can you run a CDM on with a larger test size like 16GiB or 32GiB?