Question my SSD is horrible for sequential writes

[DigDog]

i'm thinking that SSD stands for redacted because i moved a 6Gb file - all made up of 3/4/5Mb songs from one SSD to the other and it's taken 10 minutes to do the operation.

I admit my fault that i have low-tier, cheap SSDs but jesus, i didnt think it would be this bad. Sequential writes in the average of 3MB/s.

Currently have an ADATA SU800 (500Gb) and a Kingston SDSSDP128G 128Gb. Was moving from the 500 to the 128, which was about 80% full. Who's at fault here, the 128Gb? Are both bad? (i also have an even crappier 60Gb disk that isn't really being used)

Would things get substantially better if i splashed for a Samsung 860 EVO ?

Is the ADATA comparable to the 860 EVO or would the EVO be substantially better?

This kind of language is never acceptable
in the tech forums.

AT Mod Usandthem

 

[VirtualLarry]

Buy crap SSDs, get crap performance.Pretty simple. Stick to MLC (Samsung 850 Pro) drives if you want consistently-high seq. write performance.

 

[DigDog]

*yeah* but on paper, the ones i buy are nearly as good as the EVOs. Turns out that their benchmarks lied, crazy uh ?
I get reads of 400MB/s ... for a couple seconds. Then it plummets .. 70MB ... 30MB .. 8MB .. 3MB ..

 

[ArisVer]

I have a KingSpec(!) 1TB SATA SSD as my secondary drive on my main. Until 75-80% it used to write at full speed (about 450MB/s iirc, all 50GB of copy), but afterwards it started crawling. After the first few GB it goes down to even 15MB/s, so I try to keep it under 75% now. My guess is that your drive was "full" enough.

 

[Muadib]

Here's a test with my Gen4 Corsair MP600 with 29% of room left. Still nice, but it's lost a step.

-----------------------------------------------------------------------
CrystalDiskMark 6.0.2 x64 (C) 2007-2018 hiyohiyo
Crystal Dew World : https://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes

Sequential Read (Q= 32,T= 1) : 4979.928 MB/s
Sequential Write (Q= 32,T= 1) : 4260.102 MB/s
Random Read 4KiB (Q= 8,T= 8) : 1407.845 MB/s [ 343712.2 IOPS]
Random Write 4KiB (Q= 8,T= 8) : 1030.687 MB/s [ 251632.6 IOPS]
Random Read 4KiB (Q= 32,T= 1) : 345.178 MB/s [ 84272.0 IOPS]
Random Write 4KiB (Q= 32,T= 1) : 250.090 MB/s [ 61057.1 IOPS]
Random Read 4KiB (Q= 1,T= 1) : 59.856 MB/s [ 14613.3 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 201.256 MB/s [ 49134.8 IOPS]

Test : 1024 MiB [C: 29.3% (272.8/930.2 GiB)] (x5) [Interval=5 sec]
Date : 2020/07/01 11:05:20
OS : Windows 10 Professional [10.0 Build 19041] (x64)

 

[DigDog]

lol, i wish.

 

[IntelUser2000]

3MB/s is horrible. I think that's in the level of transfer rates for USB sticks.

 

[thecoolnessrune]

Long ago I remember Servethehome, AnandTech, and similar making mention that if you want peak performance to be maintained, you need to maintain 75% free, including spare area, and over the years that's tended to hold true. It becomes even more true when you have a cut down SSD that doesn't have fully populated channels. Or if you've got one of those old SandForce SSD's with Asynchronous NAND.

 

[Shmee]

I seem to understand the SU800 is an older but decent MLC drive, but that kingston that is almost full is likely to be the culprit.

 

[VirtualLarry]

No, Adata SU800 Ultimate was one of the very first 3D TLC (Intel/Micron 32-layer NAND) drives, which used three NAND dies, for 144GB of space, on their "128GB" drive model, due to bit layout.

It was very fast for reads, and fast for random writes hitting the SLC cache, but exhaust the SLC cache, and writes slowed down to like 30MB/sec instead of 500MB/sec. It was unfortunately, quite noticable, and to add insult to injury, way slower than just a decent mechanical HDD (CMR, not SMR).

 

[Shmee]

No, Adata SU800 Ultimate was one of the very first 3D TLC (Intel/Micron 32-layer NAND) drives, which used three NAND dies, for 144GB of space, on their "128GB" drive model, due to bit layout.

It was very fast for reads, and fast for random writes hitting the SLC cache, but exhaust the SLC cache, and writes slowed down to like 30MB/sec instead of 500MB/sec. It was unfortunately, quite noticable, and to add insult to injury, way slower than just a decent mechanical HDD (CMR, not SMR).

Oh ya you are right, SU 800 uses 3D TLC. I have a 1TB one in my mATX rig for additional storage.