SN550 - Why it uses 512B sector instead of 4096?

Same here. But seems no valueable advantage in using 4K size …

Some information:

# Gain Optimal Performance by Changes to SSD Physical Sector Size

Lets have a look at longevity whilst in 4096 format.

When I read out S.M.A.R.T. data, it says that it's in 4096 LBA size format.
> Supported LBA Sizes (NSID 0x1)
> Id Fmt  Data  Metadt  Rel_Perf
>  0 -     512       0         2
>  1 +    4096       0         1

Furthermore I can find S.M.A.R.T. attribute:
Data Units Written: 6.383.576
(as reported by WD SSD Dashboard (Win 10) as well as by smartctl (Linux))

That should mean that 6.383.576 units of size 4096 Bytes have been written to disk?
6.383.576 x 4096 = 26147127296 B ~ 26 GB or 24 GiB

This number can’t be right, because Win 10 installation alone is size 74 GiB.

Where do I go wrong in my calculation? Is there a WD definition of the S.M.A.R.T. attributes that the SN550 provides to be found anywhere???

Besides that does it mean that SSD-wear will be higher, now that LBA size format is 4096B instead of 512B?

The opening post stated that

Is it not just the other way around, lager sector sizes cause 8x as much data written (especially for small files <= 512B)?

I think the main reason for the drives shipping with 512B format is compatibility. Notably, cloning partitions or restoring drive images from a 512B hard disk/SSD to a 4096B format SSD is not possible.

Personally I changed my SN850 to 4096B format using nvme-cli and saw a very slight performance improvement under Windows/NTFS (which I had to reinstall from scratch, no restoring possible from existing images), but I am still not sure if I made the right choice.

It’s supposed to be in thousands units of 512 bytes, regardless of whether the drive has been set to 4096B logical sectors.

6383576 * 1000 * 512 = 3268390912000 bytes (3.26 TB or 2.97 TiB)

I don’t think wear will be any higher since internally the drive already uses a 4096B “physical” sector format, but by setting it also to 4096B “logical” sectors (if this physical/logical distinction makes any sense on SSDs) the drive will not have to do the “translation” from 512B sectors and likely have less processing overhead, which probably why performance slightly improves (or at least, it’s supposed to).

Also see: Advanced Format - Wikipedia

For sector size, it depends on what the controller and NAND is optimized for. For an SSD, while a range of sector sized can be used that can even maintain alignment, the issue is that any deviation from the design spec, will reduce performance, or at best have a slight gain in one area and a large penalty in other areas.

A larger size than needed, will cause an IOPS penalty on the drive, and harm tasks such as boot speeds and game load times, as well as other IO intensive workloads, as well as increase storage use for smaller files.

A sector size that is too small, will cause more overhead and you will get a drop in linear as well as random reads/ writes, in addition to an IOPS penalty.

Ultimately, an SSD will have its own back end way of managing the NAND, but it still needs data to be fed to it in the most efficient way possible to reach advertised performance (provided there are no major firmware flaws like with the SN750).

@Volatile.Memory @Razor512 thanks for your replies.
I found optimal throughput not at 4KB, but at 16 KB block size (whilst in 4KiB advanced format).
So it seems that the drive works internally with 16 KB physical blocks.
(See my test results)

Is there any documentation from WD available?

Is there a way to utilize this finding to gain optimal performance?

Sadly the SSD makers tend to not release detailed info about the inner workings of their controllers and firmware since that is often how they optimize performance. For example, it is common to see a ton of different SSD makers using the same Phison branded controller as well as the same model of NAND, but end up with a wide range of benchmarking results.

Usually they will optimize for default partition setups, e.g., how will a fresh windows install partition the drive by default.
Their goal is to minimize how much space is wasted, while maximizing performance across as many workloads as possible.

Sadly without the detailed information, someone would need to run a large range of tests, along the lines of this The Western Digital WD Black SN850 Review: A Very Fast PCIe 4.0 SSD on both the SLC and TLC (or QLC in the case of the latest SN550 vs TLC in last year’s model) to see which sector size performs best overall.
Some sizes could potentially increase QD1 single threaded operations in a linear read and write scenario but harm performance in mixed and QD2+ workloads.

At a server or database level, some people may choose to optimize specifically for a single workload if they are dedicating that equipment just to that workload, but for a general use PC, it will at some point experience a little bit of everything, and that is what consumer SSD makers try to aim for.

Or in the worst case, some companies can make drives where the controller does very little and you are closer to just having NAND attached to the PCIe bus, thus the workload needs to be tailored to the behavior of the NAND for best performance. Fastest Storage Possible: SPDK & P5800X Are More Like Memory than NVMe - YouTube

compared those both settings (512 vs 4096)

512:

4096:

Looking at those benchmark, you can see theres no difference in performance.

Was using GitHub - earlephilhower/ezfio: Simple NVME/SAS/SATA SSD test framework for Linux and Windows with “–fast-precondition”

using it with “long-preconditioning” it looks like this:

(512, which took arround 6 days to complete)

thank me later.

@blonkel
Wow, great graphs.
Thanks for your efforts.
I also found optimal throughput at 16 KB block size.

What does “long-preconditioning” mean exactly?

Don’t I see slightly improved performance in Sustained Sequential Reads for 4096 around the 8192 block size mark formate?

So we can conclude that there’s almost no performance increase to be had at 4096 format. So the only remaining benefits will be the improved ECC and less space consumed for it. Will 4096 format improve longevity?

Did you look into cooling the drive? Have a look over here?

Did you test with one of those newer drives of the same SN550 model, that suffer from slow NAND, which shows when they run out of 12GB of cache (at the 32768 mark) (see also there).